1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S --riscv-v-register-bit-width-lmul=1 | FileCheck %s -check-prefix=LMUL1 3; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S --riscv-v-register-bit-width-lmul=2 | FileCheck %s -check-prefix=LMUL2 4; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S --riscv-v-register-bit-width-lmul=4 | FileCheck %s -check-prefix=LMUL4 5; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S --riscv-v-register-bit-width-lmul=8 | FileCheck %s -check-prefix=LMUL8 6; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S | FileCheck %s -check-prefix=LMUL2 7 8define void @load_store(ptr %p) { 9; LMUL1-LABEL: @load_store( 10; LMUL1-NEXT: entry: 11; LMUL1-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 12; LMUL1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP0]] 13; LMUL1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 14; LMUL1: vector.ph: 15; LMUL1-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64() 16; LMUL1-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP1]] 17; LMUL1-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] 18; LMUL1-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64() 19; LMUL1-NEXT: br label [[VECTOR_BODY:%.*]] 20; LMUL1: vector.body: 21; LMUL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 22; LMUL1-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0 23; LMUL1-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[P:%.*]], i64 [[TMP2]] 24; LMUL1-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP3]], i32 0 25; LMUL1-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 1 x i64>, ptr [[TMP4]], align 8 26; LMUL1-NEXT: [[TMP5:%.*]] = add <vscale x 1 x i64> [[WIDE_LOAD]], splat (i64 1) 27; LMUL1-NEXT: store <vscale x 1 x i64> [[TMP5]], ptr [[TMP4]], align 8 28; LMUL1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]] 29; LMUL1-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 30; LMUL1-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 31; LMUL1: middle.block: 32; LMUL1-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] 33; LMUL1-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 34; LMUL1: scalar.ph: 35; LMUL1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 36; LMUL1-NEXT: br label [[FOR_BODY:%.*]] 37; LMUL1: for.body: 38; LMUL1-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 39; LMUL1-NEXT: [[Q:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[IV]] 40; LMUL1-NEXT: [[V:%.*]] = load i64, ptr [[Q]], align 8 41; LMUL1-NEXT: [[W:%.*]] = add i64 [[V]], 1 42; LMUL1-NEXT: store i64 [[W]], ptr [[Q]], align 8 43; LMUL1-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 44; LMUL1-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 45; LMUL1-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 46; LMUL1: for.end: 47; LMUL1-NEXT: ret void 48; 49; LMUL2-LABEL: @load_store( 50; LMUL2-NEXT: entry: 51; LMUL2-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 52; LMUL2-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 2 53; LMUL2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP1]] 54; LMUL2-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 55; LMUL2: vector.ph: 56; LMUL2-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 57; LMUL2-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 2 58; LMUL2-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] 59; LMUL2-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] 60; LMUL2-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64() 61; LMUL2-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 2 62; LMUL2-NEXT: br label [[VECTOR_BODY:%.*]] 63; LMUL2: vector.body: 64; LMUL2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 65; LMUL2-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 66; LMUL2-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[P:%.*]], i64 [[TMP4]] 67; LMUL2-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP5]], i32 0 68; LMUL2-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i64>, ptr [[TMP6]], align 8 69; LMUL2-NEXT: [[TMP7:%.*]] = add <vscale x 2 x i64> [[WIDE_LOAD]], splat (i64 1) 70; LMUL2-NEXT: store <vscale x 2 x i64> [[TMP7]], ptr [[TMP6]], align 8 71; LMUL2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP9]] 72; LMUL2-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 73; LMUL2-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 74; LMUL2: middle.block: 75; LMUL2-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] 76; LMUL2-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 77; LMUL2: scalar.ph: 78; LMUL2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 79; LMUL2-NEXT: br label [[FOR_BODY:%.*]] 80; LMUL2: for.body: 81; LMUL2-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 82; LMUL2-NEXT: [[Q:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[IV]] 83; LMUL2-NEXT: [[V:%.*]] = load i64, ptr [[Q]], align 8 84; LMUL2-NEXT: [[W:%.*]] = add i64 [[V]], 1 85; LMUL2-NEXT: store i64 [[W]], ptr [[Q]], align 8 86; LMUL2-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 87; LMUL2-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 88; LMUL2-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 89; LMUL2: for.end: 90; LMUL2-NEXT: ret void 91; 92; LMUL4-LABEL: @load_store( 93; LMUL4-NEXT: entry: 94; LMUL4-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 95; LMUL4-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 4 96; LMUL4-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP1]] 97; LMUL4-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 98; LMUL4: vector.ph: 99; LMUL4-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 100; LMUL4-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 4 101; LMUL4-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] 102; LMUL4-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] 103; LMUL4-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64() 104; LMUL4-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 4 105; LMUL4-NEXT: br label [[VECTOR_BODY:%.*]] 106; LMUL4: vector.body: 107; LMUL4-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 108; LMUL4-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 109; LMUL4-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[P:%.*]], i64 [[TMP4]] 110; LMUL4-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP5]], i32 0 111; LMUL4-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i64>, ptr [[TMP6]], align 8 112; LMUL4-NEXT: [[TMP7:%.*]] = add <vscale x 4 x i64> [[WIDE_LOAD]], splat (i64 1) 113; LMUL4-NEXT: store <vscale x 4 x i64> [[TMP7]], ptr [[TMP6]], align 8 114; LMUL4-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP9]] 115; LMUL4-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 116; LMUL4-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 117; LMUL4: middle.block: 118; LMUL4-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] 119; LMUL4-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 120; LMUL4: scalar.ph: 121; LMUL4-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 122; LMUL4-NEXT: br label [[FOR_BODY:%.*]] 123; LMUL4: for.body: 124; LMUL4-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 125; LMUL4-NEXT: [[Q:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[IV]] 126; LMUL4-NEXT: [[V:%.*]] = load i64, ptr [[Q]], align 8 127; LMUL4-NEXT: [[W:%.*]] = add i64 [[V]], 1 128; LMUL4-NEXT: store i64 [[W]], ptr [[Q]], align 8 129; LMUL4-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 130; LMUL4-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 131; LMUL4-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 132; LMUL4: for.end: 133; LMUL4-NEXT: ret void 134; 135; LMUL8-LABEL: @load_store( 136; LMUL8-NEXT: entry: 137; LMUL8-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 138; LMUL8-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 8 139; LMUL8-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP1]] 140; LMUL8-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 141; LMUL8: vector.ph: 142; LMUL8-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 143; LMUL8-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 8 144; LMUL8-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] 145; LMUL8-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] 146; LMUL8-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64() 147; LMUL8-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8 148; LMUL8-NEXT: br label [[VECTOR_BODY:%.*]] 149; LMUL8: vector.body: 150; LMUL8-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 151; LMUL8-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 152; LMUL8-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[P:%.*]], i64 [[TMP4]] 153; LMUL8-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP5]], i32 0 154; LMUL8-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i64>, ptr [[TMP6]], align 8 155; LMUL8-NEXT: [[TMP7:%.*]] = add <vscale x 8 x i64> [[WIDE_LOAD]], splat (i64 1) 156; LMUL8-NEXT: store <vscale x 8 x i64> [[TMP7]], ptr [[TMP6]], align 8 157; LMUL8-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP9]] 158; LMUL8-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 159; LMUL8-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 160; LMUL8: middle.block: 161; LMUL8-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] 162; LMUL8-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 163; LMUL8: scalar.ph: 164; LMUL8-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 165; LMUL8-NEXT: br label [[FOR_BODY:%.*]] 166; LMUL8: for.body: 167; LMUL8-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 168; LMUL8-NEXT: [[Q:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[IV]] 169; LMUL8-NEXT: [[V:%.*]] = load i64, ptr [[Q]], align 8 170; LMUL8-NEXT: [[W:%.*]] = add i64 [[V]], 1 171; LMUL8-NEXT: store i64 [[W]], ptr [[Q]], align 8 172; LMUL8-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 173; LMUL8-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 174; LMUL8-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 175; LMUL8: for.end: 176; LMUL8-NEXT: ret void 177; 178entry: 179 br label %for.body 180 181for.body: 182 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 183 %q = getelementptr inbounds i64, ptr %p, i64 %iv 184 %v = load i64, ptr %q 185 %w = add i64 %v, 1 186 store i64 %w, ptr %q 187 %iv.next = add nuw nsw i64 %iv, 1 188 %exitcond.not = icmp eq i64 %iv.next, 1024 189 br i1 %exitcond.not, label %for.end, label %for.body 190 191for.end: 192 ret void 193} 194