1; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -S | FileCheck --check-prefix VEC4_INTERL1 %s 2; RUN: opt < %s -loop-vectorize -force-vector-interleave=2 -force-vector-width=4 -dce -instcombine -S | FileCheck --check-prefix VEC4_INTERL2 %s 3; RUN: opt < %s -loop-vectorize -force-vector-interleave=2 -force-vector-width=1 -dce -instcombine -S | FileCheck --check-prefix VEC1_INTERL2 %s 4; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=2 -dce -simplifycfg -instcombine -S | FileCheck --check-prefix VEC2_INTERL1_PRED_STORE %s 5 6@fp_inc = common global float 0.000000e+00, align 4 7 8;void fp_iv_loop1(float init, float * __restrict__ A, int N) { 9; float x = init; 10; for (int i=0; i < N; ++i) { 11; A[i] = x; 12; x -= fp_inc; 13; } 14;} 15 16; VEC4_INTERL1-LABEL: @fp_iv_loop1( 17; VEC4_INTERL1: vector.ph: 18; VEC4_INTERL1-NEXT: br label %vector.body 19; VEC4_INTERL1: vector.body: 20; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 21; VEC4_INTERL1-NEXT: [[TMP5:%.*]] = sitofp i64 [[INDEX]] to float 22; VEC4_INTERL1-NEXT: [[TMP6:%.*]] = fmul fast float %fpinc, [[TMP5]] 23; VEC4_INTERL1-NEXT: [[FP_OFFSET_IDX:%.*]] = fsub fast float %init, [[TMP6]] 24; VEC4_INTERL1-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <4 x float> undef, float [[FP_OFFSET_IDX]], i32 0 25; VEC4_INTERL1-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT3]], <4 x float> undef, <4 x i32> zeroinitializer 26; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float %fpinc, i32 0 27; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer 28; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = fmul fast <4 x float> [[DOTSPLAT]], <float 0.000000e+00, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00> 29; VEC4_INTERL1-NEXT: [[INDUCTION5:%.*]] = fsub fast <4 x float> [[BROADCAST_SPLAT4]], [[TMP7]] 30; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]] 31; VEC4_INTERL1-NEXT: [[TMP9:%.*]] = bitcast float* [[TMP8]] to <4 x float>* 32; VEC4_INTERL1-NEXT: store <4 x float> [[INDUCTION5]], <4 x float>* [[TMP9]], align 4 33; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 34; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body 35 36; VEC4_INTERL2-LABEL: @fp_iv_loop1( 37; VEC4_INTERL2: vector.ph: 38; VEC4_INTERL2-NEXT: br label %vector.body 39; VEC4_INTERL2: vector.body: 40; VEC4_INTERL2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 41; VEC4_INTERL2-NEXT: [[TMP5:%.*]] = sitofp i64 [[INDEX]] to float 42; VEC4_INTERL2-NEXT: [[TMP6:%.*]] = fmul fast float %fpinc, [[TMP5]] 43; VEC4_INTERL2-NEXT: [[FP_OFFSET_IDX:%.*]] = fsub fast float %init, [[TMP6]] 44; VEC4_INTERL2-NEXT: [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <4 x float> undef, float [[FP_OFFSET_IDX]], i32 0 45; VEC4_INTERL2-NEXT: [[BROADCAST_SPLAT5:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT4]], <4 x float> undef, <4 x i32> zeroinitializer 46; VEC4_INTERL2-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float %fpinc, i32 0 47; VEC4_INTERL2-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer 48; VEC4_INTERL2-NEXT: [[TMP7:%.*]] = fmul fast <4 x float> [[DOTSPLAT]], <float 0.000000e+00, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00> 49; VEC4_INTERL2-NEXT: [[INDUCTION6:%.*]] = fsub fast <4 x float> [[BROADCAST_SPLAT5]], [[TMP7]] 50; VEC4_INTERL2-NEXT: [[TMP8:%.*]] = fmul fast <4 x float> [[DOTSPLAT]], <float 4.000000e+00, float 5.000000e+00, float 6.000000e+00, float 7.000000e+00> 51; VEC4_INTERL2-NEXT: [[INDUCTION9:%.*]] = fsub fast <4 x float> [[BROADCAST_SPLAT5]], [[TMP8]] 52; VEC4_INTERL2-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]] 53; VEC4_INTERL2-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to <4 x float>* 54; VEC4_INTERL2-NEXT: store <4 x float> [[INDUCTION6]], <4 x float>* [[TMP10]], align 4 55; VEC4_INTERL2-NEXT: [[TMP11:%.*]] = getelementptr float, float* [[TMP9]], i64 4 56; VEC4_INTERL2-NEXT: [[TMP12:%.*]] = bitcast float* [[TMP11]] to <4 x float>* 57; VEC4_INTERL2-NEXT: store <4 x float> [[INDUCTION9]], <4 x float>* [[TMP12]], align 4 58; VEC4_INTERL2-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 8 59; VEC4_INTERL2: br i1 {{.*}}, label %middle.block, label %vector.body 60 61; VEC1_INTERL2-LABEL: @fp_iv_loop1( 62; VEC1_INTERL2: vector.ph: 63; VEC1_INTERL2-NEXT: br label %vector.body 64; VEC1_INTERL2: vector.body: 65; VEC1_INTERL2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 66; VEC1_INTERL2-NEXT: [[INDUCTION2:%.*]] = or i64 [[INDEX]], 1 67; VEC1_INTERL2-NEXT: [[TMP6:%.*]] = sitofp i64 [[INDEX]] to float 68; VEC1_INTERL2-NEXT: [[TMP7:%.*]] = fmul fast float %fpinc, [[TMP6]] 69; VEC1_INTERL2-NEXT: [[FP_OFFSET_IDX:%.*]] = fsub fast float %init, [[TMP7]] 70; VEC1_INTERL2-NEXT: [[TMP8:%.*]] = fsub fast float [[FP_OFFSET_IDX]], %fpinc 71; VEC1_INTERL2-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]] 72; VEC1_INTERL2-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDUCTION2]] 73; VEC1_INTERL2-NEXT: store float [[FP_OFFSET_IDX]], float* [[TMP9]], align 4 74; VEC1_INTERL2-NEXT: store float [[TMP8]], float* [[TMP10]], align 4 75; VEC1_INTERL2-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2 76; VEC1_INTERL2: br i1 {{.*}}, label %middle.block, label %vector.body 77 78define void @fp_iv_loop1(float %init, float* noalias nocapture %A, i32 %N) #1 { 79entry: 80 %cmp4 = icmp sgt i32 %N, 0 81 br i1 %cmp4, label %for.body.lr.ph, label %for.end 82 83for.body.lr.ph: ; preds = %entry 84 %fpinc = load float, float* @fp_inc, align 4 85 br label %for.body 86 87for.body: ; preds = %for.body, %for.body.lr.ph 88 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ] 89 %x.05 = phi float [ %init, %for.body.lr.ph ], [ %add, %for.body ] 90 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv 91 store float %x.05, float* %arrayidx, align 4 92 %add = fsub fast float %x.05, %fpinc 93 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 94 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 95 %exitcond = icmp eq i32 %lftr.wideiv, %N 96 br i1 %exitcond, label %for.end.loopexit, label %for.body 97 98for.end.loopexit: ; preds = %for.body 99 br label %for.end 100 101for.end: ; preds = %for.end.loopexit, %entry 102 ret void 103} 104 105;void fp_iv_loop2(float init, float * __restrict__ A, int N) { 106; float x = init; 107; for (int i=0; i < N; ++i) { 108; A[i] = x; 109; x += 0.5; 110; } 111;} 112 113; VEC4_INTERL1-LABEL: @fp_iv_loop2( 114; VEC4_INTERL1: vector.ph: 115; VEC4_INTERL1-NEXT: br label %vector.body 116; VEC4_INTERL1: vector.body: 117; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 118; VEC4_INTERL1-NEXT: [[TMP5:%.*]] = sitofp i64 [[INDEX]] to float 119; VEC4_INTERL1-NEXT: [[TMP6:%.*]] = fmul fast float [[TMP5]], 5.000000e-01 120; VEC4_INTERL1-NEXT: [[FP_OFFSET_IDX:%.*]] = fadd fast float [[TMP6]], %init 121; VEC4_INTERL1-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <4 x float> undef, float [[FP_OFFSET_IDX]], i32 0 122; VEC4_INTERL1-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT3]], <4 x float> undef, <4 x i32> zeroinitializer 123; VEC4_INTERL1-NEXT: [[INDUCTION5:%.*]] = fadd fast <4 x float> [[BROADCAST_SPLAT4]], <float 0.000000e+00, float 5.000000e-01, float 1.000000e+00, float 1.500000e+00> 124; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]] 125; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = bitcast float* [[TMP7]] to <4 x float>* 126; VEC4_INTERL1-NEXT: store <4 x float> [[INDUCTION5]], <4 x float>* [[TMP8]], align 4 127; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 128; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body 129 130define void @fp_iv_loop2(float %init, float* noalias nocapture %A, i32 %N) #0 { 131entry: 132 %cmp4 = icmp sgt i32 %N, 0 133 br i1 %cmp4, label %for.body.preheader, label %for.end 134 135for.body.preheader: ; preds = %entry 136 br label %for.body 137 138for.body: ; preds = %for.body.preheader, %for.body 139 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ] 140 %x.06 = phi float [ %conv1, %for.body ], [ %init, %for.body.preheader ] 141 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv 142 store float %x.06, float* %arrayidx, align 4 143 %conv1 = fadd fast float %x.06, 5.000000e-01 144 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 145 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 146 %exitcond = icmp eq i32 %lftr.wideiv, %N 147 br i1 %exitcond, label %for.end.loopexit, label %for.body 148 149for.end.loopexit: ; preds = %for.body 150 br label %for.end 151 152for.end: ; preds = %for.end.loopexit, %entry 153 ret void 154} 155 156;void fp_iv_loop3(float init, float * __restrict__ A, float * __restrict__ B, float * __restrict__ C, int N) { 157; int i = 0; 158; float x = init; 159; float y = 0.1; 160; for (; i < N; ++i) { 161; A[i] = x; 162; x += fp_inc; 163; y -= 0.5; 164; B[i] = x + y; 165; C[i] = y; 166; } 167;} 168 169; VEC4_INTERL1-LABEL: @fp_iv_loop3( 170; VEC4_INTERL1: for.body.lr.ph: 171; VEC4_INTERL1: [[TMP0:%.*]] = load float, float* @fp_inc, align 4 172; VEC4_INTERL1: vector.ph: 173; VEC4_INTERL1-NEXT: [[BROADCAST_SPLATINSERT14:%.*]] = insertelement <4 x float> undef, float [[TMP0]], i32 0 174; VEC4_INTERL1-NEXT: [[BROADCAST_SPLAT15:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT14]], <4 x float> undef, <4 x i32> zeroinitializer 175; VEC4_INTERL1-NEXT: br label %vector.body 176; VEC4_INTERL1: vector.body: 177; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 178; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = sitofp i64 [[INDEX]] to float 179; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = fmul fast float [[TMP7]], -5.000000e-01 180; VEC4_INTERL1-NEXT: [[FP_OFFSET_IDX:%.*]] = fadd fast float [[TMP8]], 0x3FB99999A0000000 181; VEC4_INTERL1-NEXT: [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <4 x float> undef, float [[FP_OFFSET_IDX]], i32 0 182; VEC4_INTERL1-NEXT: [[BROADCAST_SPLAT7:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT6]], <4 x float> undef, <4 x i32> zeroinitializer 183; VEC4_INTERL1-NEXT: [[TMP9:%.*]] = sitofp i64 [[INDEX]] to float 184; VEC4_INTERL1-NEXT: [[TMP10:%.*]] = fmul fast float [[TMP0]], [[TMP9]] 185; VEC4_INTERL1-NEXT: [[FP_OFFSET_IDX10:%.*]] = fadd fast float [[TMP10]], %init 186; VEC4_INTERL1-NEXT: [[BROADCAST_SPLATINSERT11:%.*]] = insertelement <4 x float> undef, float [[FP_OFFSET_IDX10]], i32 0 187; VEC4_INTERL1-NEXT: [[BROADCAST_SPLAT12:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT11]], <4 x float> undef, <4 x i32> zeroinitializer 188; VEC4_INTERL1-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float [[TMP0]], i32 0 189; VEC4_INTERL1-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer 190; VEC4_INTERL1-NEXT: [[TMP11:%.*]] = fmul fast <4 x float> [[DOTSPLAT]], <float 0.000000e+00, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00> 191; VEC4_INTERL1-NEXT: [[INDUCTION13:%.*]] = fadd fast <4 x float> [[BROADCAST_SPLAT12]], [[TMP11]] 192; VEC4_INTERL1-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]] 193; VEC4_INTERL1-NEXT: [[TMP13:%.*]] = bitcast float* [[TMP12]] to <4 x float>* 194; VEC4_INTERL1-NEXT: store <4 x float> [[INDUCTION13]], <4 x float>* [[TMP13]], align 4 195; VEC4_INTERL1-NEXT: [[TMP14:%.*]] = fadd fast <4 x float> [[INDUCTION13]], [[BROADCAST_SPLAT15]] 196; VEC4_INTERL1-NEXT: [[TMP15:%.*]] = fadd fast <4 x float> [[BROADCAST_SPLAT7]], <float -5.000000e-01, float -1.000000e+00, float -1.500000e+00, float -2.000000e+00> 197; VEC4_INTERL1-NEXT: [[TMP16:%.*]] = fadd fast <4 x float> [[TMP15]], [[TMP14]] 198; VEC4_INTERL1-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float* %B, i64 [[INDEX]] 199; VEC4_INTERL1-NEXT: [[TMP18:%.*]] = bitcast float* [[TMP17]] to <4 x float>* 200; VEC4_INTERL1-NEXT: store <4 x float> [[TMP16]], <4 x float>* [[TMP18]], align 4 201; VEC4_INTERL1-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float* %C, i64 [[INDEX]] 202; VEC4_INTERL1-NEXT: [[TMP20:%.*]] = bitcast float* [[TMP19]] to <4 x float>* 203; VEC4_INTERL1-NEXT: store <4 x float> [[TMP15]], <4 x float>* [[TMP20]], align 4 204; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 205; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body 206 207define void @fp_iv_loop3(float %init, float* noalias nocapture %A, float* noalias nocapture %B, float* noalias nocapture %C, i32 %N) #1 { 208entry: 209 %cmp9 = icmp sgt i32 %N, 0 210 br i1 %cmp9, label %for.body.lr.ph, label %for.end 211 212for.body.lr.ph: ; preds = %entry 213 %0 = load float, float* @fp_inc, align 4 214 br label %for.body 215 216for.body: ; preds = %for.body, %for.body.lr.ph 217 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ] 218 %y.012 = phi float [ 0x3FB99999A0000000, %for.body.lr.ph ], [ %conv1, %for.body ] 219 %x.011 = phi float [ %init, %for.body.lr.ph ], [ %add, %for.body ] 220 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv 221 store float %x.011, float* %arrayidx, align 4 222 %add = fadd fast float %x.011, %0 223 %conv1 = fadd fast float %y.012, -5.000000e-01 224 %add2 = fadd fast float %conv1, %add 225 %arrayidx4 = getelementptr inbounds float, float* %B, i64 %indvars.iv 226 store float %add2, float* %arrayidx4, align 4 227 %arrayidx6 = getelementptr inbounds float, float* %C, i64 %indvars.iv 228 store float %conv1, float* %arrayidx6, align 4 229 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 230 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 231 %exitcond = icmp eq i32 %lftr.wideiv, %N 232 br i1 %exitcond, label %for.end.loopexit, label %for.body 233 234for.end.loopexit: 235 br label %for.end 236 237for.end: 238 ret void 239} 240 241; Start and step values are constants. There is no 'fmul' operation in this case 242;void fp_iv_loop4(float * __restrict__ A, int N) { 243; float x = 1.0; 244; for (int i=0; i < N; ++i) { 245; A[i] = x; 246; x += 0.5; 247; } 248;} 249 250; VEC4_INTERL1-LABEL: @fp_iv_loop4( 251; VEC4_INTERL1: vector.ph: 252; VEC4_INTERL1-NEXT: br label %vector.body 253; VEC4_INTERL1: vector.body: 254; VEC4_INTERL1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 255; VEC4_INTERL1-NEXT: [[TMP5:%.*]] = sitofp i64 [[INDEX]] to float 256; VEC4_INTERL1-NEXT: [[TMP6:%.*]] = fmul fast float [[TMP5]], 5.000000e-01 257; VEC4_INTERL1-NEXT: [[FP_OFFSET_IDX:%.*]] = fadd fast float [[TMP6]], 1.000000e+00 258; VEC4_INTERL1-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <4 x float> undef, float [[FP_OFFSET_IDX]], i32 0 259; VEC4_INTERL1-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT3]], <4 x float> undef, <4 x i32> zeroinitializer 260; VEC4_INTERL1-NEXT: [[INDUCTION5:%.*]] = fadd fast <4 x float> [[BROADCAST_SPLAT4]], <float 0.000000e+00, float 5.000000e-01, float 1.000000e+00, float 1.500000e+00> 261; VEC4_INTERL1-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]] 262; VEC4_INTERL1-NEXT: [[TMP8:%.*]] = bitcast float* [[TMP7]] to <4 x float>* 263; VEC4_INTERL1-NEXT: store <4 x float> [[INDUCTION5]], <4 x float>* [[TMP8]], align 4 264; VEC4_INTERL1-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4 265; VEC4_INTERL1: br i1 {{.*}}, label %middle.block, label %vector.body 266 267define void @fp_iv_loop4(float* noalias nocapture %A, i32 %N) { 268entry: 269 %cmp4 = icmp sgt i32 %N, 0 270 br i1 %cmp4, label %for.body.preheader, label %for.end 271 272for.body.preheader: ; preds = %entry 273 br label %for.body 274 275for.body: ; preds = %for.body.preheader, %for.body 276 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ] 277 %x.06 = phi float [ %conv1, %for.body ], [ 1.000000e+00, %for.body.preheader ] 278 %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv 279 store float %x.06, float* %arrayidx, align 4 280 %conv1 = fadd fast float %x.06, 5.000000e-01 281 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 282 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 283 %exitcond = icmp eq i32 %lftr.wideiv, %N 284 br i1 %exitcond, label %for.end.loopexit, label %for.body 285 286for.end.loopexit: ; preds = %for.body 287 br label %for.end 288 289for.end: ; preds = %for.end.loopexit, %entry 290 ret void 291} 292 293; VEC2_INTERL1_PRED_STORE-LABEL: @non_primary_iv_float_scalar( 294; VEC2_INTERL1_PRED_STORE: vector.body: 295; VEC2_INTERL1_PRED_STORE-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE7:.*]] ], [ 0, %min.iters.checked ] 296; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP1:%.*]] = sitofp i64 [[INDEX]] to float 297; VEC2_INTERL1_PRED_STORE-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <2 x float> undef, float [[TMP1]], i32 0 298; VEC2_INTERL1_PRED_STORE-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT3]], <2 x float> undef, <2 x i32> zeroinitializer 299; VEC2_INTERL1_PRED_STORE-NEXT: [[INDUCTION5:%.*]] = fadd fast <2 x float> [[BROADCAST_SPLAT4]], <float 0.000000e+00, float 1.000000e+00> 300; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]] 301; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP3:%.*]] = bitcast float* [[TMP2]] to <2 x float>* 302; VEC2_INTERL1_PRED_STORE-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP3]], align 4 303; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP4:%.*]] = fcmp fast oeq <2 x float> [[WIDE_LOAD]], zeroinitializer 304; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0 305; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[TMP5]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]] 306; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_IF]]: 307; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP6:%.*]] = extractelement <2 x float> [[INDUCTION5]], i32 0 308; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* %A, i64 [[INDEX]] 309; VEC2_INTERL1_PRED_STORE-NEXT: store float [[TMP6]], float* [[TMP7]], align 4 310; VEC2_INTERL1_PRED_STORE-NEXT: br label %[[PRED_STORE_CONTINUE]] 311; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_CONTINUE]]: 312; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP8:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1 313; VEC2_INTERL1_PRED_STORE-NEXT: br i1 [[TMP8]], label %[[PRED_STORE_IF6:.*]], label %[[PRED_STORE_CONTINUE7]] 314; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_IF6]]: 315; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP9:%.*]] = extractelement <2 x float> [[INDUCTION5]], i32 1 316; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP10:%.*]] = or i64 [[INDEX]], 1 317; VEC2_INTERL1_PRED_STORE-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, float* %A, i64 [[TMP10]] 318; VEC2_INTERL1_PRED_STORE-NEXT: store float [[TMP9]], float* [[TMP11]], align 4 319; VEC2_INTERL1_PRED_STORE-NEXT: br label %[[PRED_STORE_CONTINUE7]] 320; VEC2_INTERL1_PRED_STORE: [[PRED_STORE_CONTINUE7]]: 321; VEC2_INTERL1_PRED_STORE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2 322; VEC2_INTERL1_PRED_STORE: br i1 {{.*}}, label %middle.block, label %vector.body 323 324define void @non_primary_iv_float_scalar(float* %A, i64 %N) { 325entry: 326 br label %for.body 327 328for.body: 329 %i = phi i64 [ %i.next, %for.inc ], [ 0, %entry ] 330 %j = phi float [ %j.next, %for.inc ], [ 0.0, %entry ] 331 %tmp0 = getelementptr inbounds float, float* %A, i64 %i 332 %tmp1 = load float, float* %tmp0, align 4 333 %tmp2 = fcmp fast oeq float %tmp1, 0.0 334 br i1 %tmp2, label %if.pred, label %for.inc 335 336if.pred: 337 store float %j, float* %tmp0, align 4 338 br label %for.inc 339 340for.inc: 341 %i.next = add nuw nsw i64 %i, 1 342 %j.next = fadd fast float %j, 1.0 343 %cond = icmp slt i64 %i.next, %N 344 br i1 %cond, label %for.body, label %for.end 345 346for.end: 347 ret void 348} 349