1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -loop-vectorize -scalable-vectorization=on -force-target-instruction-cost=1 -force-target-supports-scalable-vectors -dce -instcombine < %s -S | FileCheck %s 3 4; Test that we can add on the induction variable 5; for (long long i = 0; i < n; i++) { 6; a[i] = b[i] + i; 7; } 8; with an unroll factor (interleave count) of 2. 9 10define void @add_ind64_unrolled(i64* noalias nocapture %a, i64* noalias nocapture readonly %b, i64 %n) { 11; CHECK-LABEL: @add_ind64_unrolled( 12; CHECK-NEXT: entry: 13; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 14; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[TMP0]], 2 15; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ugt i64 [[TMP1]], [[N:%.*]] 16; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 17; CHECK: vector.ph: 18; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 19; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 2 20; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]] 21; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 22; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 23; CHECK: vector.body: 24; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 25; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.experimental.stepvector.nxv2i64() 26; CHECK-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[INDEX]], i64 0 27; CHECK-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <vscale x 2 x i64> [[DOTSPLATINSERT2]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer 28; CHECK-NEXT: [[TMP5:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT3]], [[TMP4]] 29; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64() 30; CHECK-NEXT: [[TMP7:%.*]] = shl i64 [[TMP6]], 1 31; CHECK-NEXT: [[DOTSPLATINSERT4:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP7]], i64 0 32; CHECK-NEXT: [[DOTSPLAT5:%.*]] = shufflevector <vscale x 2 x i64> [[DOTSPLATINSERT4]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer 33; CHECK-NEXT: [[TMP8:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT5]], [[TMP4]] 34; CHECK-NEXT: [[TMP9:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT3]], [[TMP8]] 35; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i64, i64* [[B:%.*]], i64 [[INDEX]] 36; CHECK-NEXT: [[TMP11:%.*]] = bitcast i64* [[TMP10]] to <vscale x 2 x i64>* 37; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP11]], align 8 38; CHECK-NEXT: [[TMP12:%.*]] = call i32 @llvm.vscale.i32() 39; CHECK-NEXT: [[TMP13:%.*]] = shl i32 [[TMP12]], 1 40; CHECK-NEXT: [[TMP14:%.*]] = sext i32 [[TMP13]] to i64 41; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i64, i64* [[TMP10]], i64 [[TMP14]] 42; CHECK-NEXT: [[TMP16:%.*]] = bitcast i64* [[TMP15]] to <vscale x 2 x i64>* 43; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <vscale x 2 x i64>, <vscale x 2 x i64>* [[TMP16]], align 8 44; CHECK-NEXT: [[TMP17:%.*]] = add nsw <vscale x 2 x i64> [[WIDE_LOAD]], [[TMP5]] 45; CHECK-NEXT: [[TMP18:%.*]] = add nsw <vscale x 2 x i64> [[WIDE_LOAD6]], [[TMP9]] 46; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i64, i64* [[A:%.*]], i64 [[INDEX]] 47; CHECK-NEXT: [[TMP20:%.*]] = bitcast i64* [[TMP19]] to <vscale x 2 x i64>* 48; CHECK-NEXT: store <vscale x 2 x i64> [[TMP17]], <vscale x 2 x i64>* [[TMP20]], align 8 49; CHECK-NEXT: [[TMP21:%.*]] = call i32 @llvm.vscale.i32() 50; CHECK-NEXT: [[TMP22:%.*]] = shl i32 [[TMP21]], 1 51; CHECK-NEXT: [[TMP23:%.*]] = sext i32 [[TMP22]] to i64 52; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i64, i64* [[TMP19]], i64 [[TMP23]] 53; CHECK-NEXT: [[TMP25:%.*]] = bitcast i64* [[TMP24]] to <vscale x 2 x i64>* 54; CHECK-NEXT: store <vscale x 2 x i64> [[TMP18]], <vscale x 2 x i64>* [[TMP25]], align 8 55; CHECK-NEXT: [[TMP26:%.*]] = call i64 @llvm.vscale.i64() 56; CHECK-NEXT: [[TMP27:%.*]] = shl i64 [[TMP26]], 2 57; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP27]] 58; CHECK-NEXT: [[TMP28:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 59; CHECK-NEXT: br i1 [[TMP28]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 60; CHECK: middle.block: 61; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 62; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 63; CHECK: scalar.ph: 64; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 65; CHECK-NEXT: br label [[FOR_BODY:%.*]] 66; CHECK: for.body: 67; CHECK-NEXT: [[I_08:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 68; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, i64* [[B]], i64 [[I_08]] 69; CHECK-NEXT: [[TMP29:%.*]] = load i64, i64* [[ARRAYIDX]], align 8 70; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP29]], [[I_08]] 71; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, i64* [[A]], i64 [[I_08]] 72; CHECK-NEXT: store i64 [[ADD]], i64* [[ARRAYIDX1]], align 8 73; CHECK-NEXT: [[INC]] = add nuw nsw i64 [[I_08]], 1 74; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]] 75; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 76; CHECK: exit: 77; CHECK-NEXT: ret void 78; 79entry: 80 br label %for.body 81 82for.body: ; preds = %entry, %for.body 83 %i.08 = phi i64 [ %inc, %for.body ], [ 0, %entry ] 84 %arrayidx = getelementptr inbounds i64, i64* %b, i64 %i.08 85 %0 = load i64, i64* %arrayidx, align 8 86 %add = add nsw i64 %0, %i.08 87 %arrayidx1 = getelementptr inbounds i64, i64* %a, i64 %i.08 88 store i64 %add, i64* %arrayidx1, align 8 89 %inc = add nuw nsw i64 %i.08, 1 90 %exitcond.not = icmp eq i64 %inc, %n 91 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !0 92 93exit: ; preds = %for.body 94 ret void 95} 96 97 98; Same as above, except we test with a vectorisation factor of (1, scalable) 99 100define void @add_ind64_unrolled_nxv1i64(i64* noalias nocapture %a, i64* noalias nocapture readonly %b, i64 %n) { 101; CHECK-LABEL: @add_ind64_unrolled_nxv1i64( 102; CHECK-NEXT: entry: 103; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 104; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[TMP0]], 1 105; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ugt i64 [[TMP1]], [[N:%.*]] 106; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 107; CHECK: vector.ph: 108; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 109; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 1 110; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]] 111; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 112; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 113; CHECK: vector.body: 114; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 115; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 1 x i64> @llvm.experimental.stepvector.nxv1i64() 116; CHECK-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[INDEX]], i64 0 117; CHECK-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT2]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer 118; CHECK-NEXT: [[TMP5:%.*]] = add <vscale x 1 x i64> [[DOTSPLAT3]], [[TMP4]] 119; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64() 120; CHECK-NEXT: [[DOTSPLATINSERT4:%.*]] = insertelement <vscale x 1 x i64> poison, i64 [[TMP6]], i64 0 121; CHECK-NEXT: [[DOTSPLAT5:%.*]] = shufflevector <vscale x 1 x i64> [[DOTSPLATINSERT4]], <vscale x 1 x i64> poison, <vscale x 1 x i32> zeroinitializer 122; CHECK-NEXT: [[TMP7:%.*]] = add <vscale x 1 x i64> [[DOTSPLAT5]], [[TMP4]] 123; CHECK-NEXT: [[TMP8:%.*]] = add <vscale x 1 x i64> [[DOTSPLAT3]], [[TMP7]] 124; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i64, i64* [[B:%.*]], i64 [[INDEX]] 125; CHECK-NEXT: [[TMP10:%.*]] = bitcast i64* [[TMP9]] to <vscale x 1 x i64>* 126; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 1 x i64>, <vscale x 1 x i64>* [[TMP10]], align 8 127; CHECK-NEXT: [[TMP11:%.*]] = call i32 @llvm.vscale.i32() 128; CHECK-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64 129; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i64, i64* [[TMP9]], i64 [[TMP12]] 130; CHECK-NEXT: [[TMP14:%.*]] = bitcast i64* [[TMP13]] to <vscale x 1 x i64>* 131; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <vscale x 1 x i64>, <vscale x 1 x i64>* [[TMP14]], align 8 132; CHECK-NEXT: [[TMP15:%.*]] = add nsw <vscale x 1 x i64> [[WIDE_LOAD]], [[TMP5]] 133; CHECK-NEXT: [[TMP16:%.*]] = add nsw <vscale x 1 x i64> [[WIDE_LOAD6]], [[TMP8]] 134; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i64, i64* [[A:%.*]], i64 [[INDEX]] 135; CHECK-NEXT: [[TMP18:%.*]] = bitcast i64* [[TMP17]] to <vscale x 1 x i64>* 136; CHECK-NEXT: store <vscale x 1 x i64> [[TMP15]], <vscale x 1 x i64>* [[TMP18]], align 8 137; CHECK-NEXT: [[TMP19:%.*]] = call i32 @llvm.vscale.i32() 138; CHECK-NEXT: [[TMP20:%.*]] = sext i32 [[TMP19]] to i64 139; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i64, i64* [[TMP17]], i64 [[TMP20]] 140; CHECK-NEXT: [[TMP22:%.*]] = bitcast i64* [[TMP21]] to <vscale x 1 x i64>* 141; CHECK-NEXT: store <vscale x 1 x i64> [[TMP16]], <vscale x 1 x i64>* [[TMP22]], align 8 142; CHECK-NEXT: [[TMP23:%.*]] = call i64 @llvm.vscale.i64() 143; CHECK-NEXT: [[TMP24:%.*]] = shl i64 [[TMP23]], 1 144; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP24]] 145; CHECK-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 146; CHECK-NEXT: br i1 [[TMP25]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 147; CHECK: middle.block: 148; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 149; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 150; CHECK: scalar.ph: 151; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 152; CHECK-NEXT: br label [[FOR_BODY:%.*]] 153; CHECK: for.body: 154; CHECK-NEXT: [[I_08:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 155; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, i64* [[B]], i64 [[I_08]] 156; CHECK-NEXT: [[TMP26:%.*]] = load i64, i64* [[ARRAYIDX]], align 8 157; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[TMP26]], [[I_08]] 158; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, i64* [[A]], i64 [[I_08]] 159; CHECK-NEXT: store i64 [[ADD]], i64* [[ARRAYIDX1]], align 8 160; CHECK-NEXT: [[INC]] = add nuw nsw i64 [[I_08]], 1 161; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]] 162; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] 163; CHECK: exit: 164; CHECK-NEXT: ret void 165; 166entry: 167 br label %for.body 168 169for.body: ; preds = %entry, %for.body 170 %i.08 = phi i64 [ %inc, %for.body ], [ 0, %entry ] 171 %arrayidx = getelementptr inbounds i64, i64* %b, i64 %i.08 172 %0 = load i64, i64* %arrayidx, align 8 173 %add = add nsw i64 %0, %i.08 174 %arrayidx1 = getelementptr inbounds i64, i64* %a, i64 %i.08 175 store i64 %add, i64* %arrayidx1, align 8 176 %inc = add nuw nsw i64 %i.08, 1 177 %exitcond.not = icmp eq i64 %inc, %n 178 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !9 179 180exit: ; preds = %for.body 181 ret void 182} 183 184 185; Test that we can vectorize a separate induction variable (not used for the branch) 186; int r = 0; 187; for (long long i = 0; i < n; i++) { 188; a[i] = r; 189; r += 2; 190; } 191; with an unroll factor (interleave count) of 1. 192 193 194define void @add_unique_ind32(i32* noalias nocapture %a, i64 %n) { 195; CHECK-LABEL: @add_unique_ind32( 196; CHECK-NEXT: entry: 197; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 198; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[TMP0]], 2 199; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ugt i64 [[TMP1]], [[N:%.*]] 200; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 201; CHECK: vector.ph: 202; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 203; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 2 204; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]] 205; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 206; CHECK-NEXT: [[CAST_CRD:%.*]] = trunc i64 [[N_VEC]] to i32 207; CHECK-NEXT: [[IND_END:%.*]] = shl i32 [[CAST_CRD]], 1 208; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 4 x i32> @llvm.experimental.stepvector.nxv4i32() 209; CHECK-NEXT: [[TMP5:%.*]] = shl <vscale x 4 x i32> [[TMP4]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 1, i32 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer) 210; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.vscale.i32() 211; CHECK-NEXT: [[TMP7:%.*]] = shl i32 [[TMP6]], 3 212; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[TMP7]], i64 0 213; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[DOTSPLATINSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer 214; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 215; CHECK: vector.body: 216; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 217; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i32> [ [[TMP5]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 218; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDEX]] 219; CHECK-NEXT: [[TMP9:%.*]] = bitcast i32* [[TMP8]] to <vscale x 4 x i32>* 220; CHECK-NEXT: store <vscale x 4 x i32> [[VEC_IND]], <vscale x 4 x i32>* [[TMP9]], align 4 221; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.vscale.i64() 222; CHECK-NEXT: [[TMP11:%.*]] = shl i64 [[TMP10]], 2 223; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP11]] 224; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i32> [[VEC_IND]], [[DOTSPLAT]] 225; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 226; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] 227; CHECK: middle.block: 228; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 229; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 230; CHECK: scalar.ph: 231; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 232; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] 233; CHECK-NEXT: br label [[FOR_BODY:%.*]] 234; CHECK: for.body: 235; CHECK-NEXT: [[I_08:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 236; CHECK-NEXT: [[R_07:%.*]] = phi i32 [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ] 237; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I_08]] 238; CHECK-NEXT: store i32 [[R_07]], i32* [[ARRAYIDX]], align 4 239; CHECK-NEXT: [[ADD]] = add nuw nsw i32 [[R_07]], 2 240; CHECK-NEXT: [[INC]] = add nuw nsw i64 [[I_08]], 1 241; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]] 242; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] 243; CHECK: exit: 244; CHECK-NEXT: ret void 245; 246entry: 247 br label %for.body 248 249for.body: ; preds = %entry, %for.body 250 %i.08 = phi i64 [ %inc, %for.body ], [ 0, %entry ] 251 %r.07 = phi i32 [ %add, %for.body ], [ 0, %entry ] 252 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %i.08 253 store i32 %r.07, i32* %arrayidx, align 4 254 %add = add nuw nsw i32 %r.07, 2 255 %inc = add nuw nsw i64 %i.08, 1 256 %exitcond.not = icmp eq i64 %inc, %n 257 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !6 258 259exit: ; preds = %for.body 260 ret void 261} 262 263 264; Test that we can vectorize a separate FP induction variable (not used for the branch) 265; float r = 0; 266; for (long long i = 0; i < n; i++) { 267; a[i] = r; 268; r += 2; 269; } 270; with an unroll factor (interleave count) of 1. 271 272define void @add_unique_indf32(float* noalias nocapture %a, i64 %n) { 273; CHECK-LABEL: @add_unique_indf32( 274; CHECK-NEXT: entry: 275; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 276; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[TMP0]], 2 277; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ugt i64 [[TMP1]], [[N:%.*]] 278; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 279; CHECK: vector.ph: 280; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 281; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 2 282; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]] 283; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 284; CHECK-NEXT: [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to float 285; CHECK-NEXT: [[TMP4:%.*]] = fmul float [[CAST_CRD]], 2.000000e+00 286; CHECK-NEXT: [[IND_END:%.*]] = fadd float [[TMP4]], 0.000000e+00 287; CHECK-NEXT: [[TMP5:%.*]] = call <vscale x 4 x i32> @llvm.experimental.stepvector.nxv4i32() 288; CHECK-NEXT: [[TMP6:%.*]] = uitofp <vscale x 4 x i32> [[TMP5]] to <vscale x 4 x float> 289; CHECK-NEXT: [[TMP7:%.*]] = fmul <vscale x 4 x float> [[TMP6]], shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 2.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer) 290; CHECK-NEXT: [[INDUCTION:%.*]] = fadd <vscale x 4 x float> [[TMP7]], zeroinitializer 291; CHECK-NEXT: [[TMP8:%.*]] = call i32 @llvm.vscale.i32() 292; CHECK-NEXT: [[TMP9:%.*]] = shl i32 [[TMP8]], 2 293; CHECK-NEXT: [[TMP10:%.*]] = uitofp i32 [[TMP9]] to float 294; CHECK-NEXT: [[TMP11:%.*]] = fmul float [[TMP10]], 2.000000e+00 295; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 4 x float> poison, float [[TMP11]], i64 0 296; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 4 x float> [[DOTSPLATINSERT]], <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer 297; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 298; CHECK: vector.body: 299; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 300; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x float> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 301; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDEX]] 302; CHECK-NEXT: [[TMP13:%.*]] = bitcast float* [[TMP12]] to <vscale x 4 x float>* 303; CHECK-NEXT: store <vscale x 4 x float> [[VEC_IND]], <vscale x 4 x float>* [[TMP13]], align 4 304; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64() 305; CHECK-NEXT: [[TMP15:%.*]] = shl i64 [[TMP14]], 2 306; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP15]] 307; CHECK-NEXT: [[VEC_IND_NEXT]] = fadd <vscale x 4 x float> [[VEC_IND]], [[DOTSPLAT]] 308; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 309; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] 310; CHECK: middle.block: 311; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 312; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 313; CHECK: scalar.ph: 314; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 315; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi float [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ] 316; CHECK-NEXT: br label [[FOR_BODY:%.*]] 317; CHECK: for.body: 318; CHECK-NEXT: [[I_08:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 319; CHECK-NEXT: [[R_07:%.*]] = phi float [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ] 320; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[I_08]] 321; CHECK-NEXT: store float [[R_07]], float* [[ARRAYIDX]], align 4 322; CHECK-NEXT: [[ADD]] = fadd float [[R_07]], 2.000000e+00 323; CHECK-NEXT: [[INC]] = add nuw nsw i64 [[I_08]], 1 324; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]] 325; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] 326; CHECK: exit: 327; CHECK-NEXT: ret void 328; 329entry: 330 br label %for.body 331 332for.body: ; preds = %entry, %for.body 333 %i.08 = phi i64 [ %inc, %for.body ], [ 0, %entry ] 334 %r.07 = phi float [ %add, %for.body ], [ 0.000000e+00, %entry ] 335 %arrayidx = getelementptr inbounds float, float* %a, i64 %i.08 336 store float %r.07, float* %arrayidx, align 4 337 %add = fadd float %r.07, 2.000000e+00 338 %inc = add nuw nsw i64 %i.08, 1 339 %exitcond.not = icmp eq i64 %inc, %n 340 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !6 341 342exit: ; preds = %for.body 343 ret void 344} 345 346!0 = distinct !{!0, !1, !2, !3, !4, !5} 347!1 = !{!"llvm.loop.mustprogress"} 348!2 = !{!"llvm.loop.vectorize.width", i32 2} 349!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true} 350!4 = !{!"llvm.loop.interleave.count", i32 2} 351!5 = !{!"llvm.loop.vectorize.enable", i1 true} 352!6 = distinct !{!6, !1, !7, !3, !8, !5} 353!7 = !{!"llvm.loop.vectorize.width", i32 4} 354!8 = !{!"llvm.loop.interleave.count", i32 1} 355!9 = distinct !{!9, !1, !10, !3, !4, !5} 356!10 = !{!"llvm.loop.vectorize.width", i32 1} 357