1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -passes=loop-vectorize \ 3; RUN: -force-tail-folding-style=data-with-evl \ 4; RUN: -prefer-predicate-over-epilogue=predicate-dont-vectorize \ 5; RUN: -mtriple=riscv64 -mattr=+v -S < %s | FileCheck --check-prefix=IF-EVL %s 6 7; RUN: opt -passes=loop-vectorize \ 8; RUN: -force-tail-folding-style=none \ 9; RUN: -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue \ 10; RUN: -mtriple=riscv64 -mattr=+v -S < %s | FileCheck --check-prefix=NO-VP %s 11 12; FIXME: interleaved accesses are not supported yet with predicated vectorization. 13define void @interleave(ptr noalias %a, ptr noalias %b, i64 %N) { 14; IF-EVL-LABEL: @interleave( 15; IF-EVL-NEXT: entry: 16; IF-EVL-NEXT: [[TMP0:%.*]] = sub i64 -1, [[N:%.*]] 17; IF-EVL-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64() 18; IF-EVL-NEXT: [[TMP2:%.*]] = mul i64 [[TMP1]], 8 19; IF-EVL-NEXT: [[TMP3:%.*]] = icmp ult i64 [[TMP0]], [[TMP2]] 20; IF-EVL-NEXT: br i1 [[TMP3]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 21; IF-EVL: vector.ph: 22; IF-EVL-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64() 23; IF-EVL-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8 24; IF-EVL-NEXT: [[TMP6:%.*]] = sub i64 [[TMP5]], 1 25; IF-EVL-NEXT: [[N_RND_UP:%.*]] = add i64 [[N]], [[TMP6]] 26; IF-EVL-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP5]] 27; IF-EVL-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]] 28; IF-EVL-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[N]], 1 29; IF-EVL-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64() 30; IF-EVL-NEXT: [[TMP8:%.*]] = mul i64 [[TMP7]], 4 31; IF-EVL-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 2 32; IF-EVL-NEXT: [[TMP10:%.*]] = call <vscale x 4 x i64> @llvm.stepvector.nxv4i64() 33; IF-EVL-NEXT: [[TMP12:%.*]] = mul <vscale x 4 x i64> [[TMP10]], splat (i64 1) 34; IF-EVL-NEXT: [[INDUCTION:%.*]] = add <vscale x 4 x i64> zeroinitializer, [[TMP12]] 35; IF-EVL-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i64> poison, i64 [[TMP8]], i64 0 36; IF-EVL-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer 37; IF-EVL-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0 38; IF-EVL-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 4 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer 39; IF-EVL-NEXT: br label [[VECTOR_BODY:%.*]] 40; IF-EVL: vector.body: 41; IF-EVL-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 42; IF-EVL-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 43; IF-EVL-NEXT: [[STEP_ADD:%.*]] = add <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] 44; IF-EVL-NEXT: [[TMP13:%.*]] = add i64 [[INDEX]], 0 45; IF-EVL-NEXT: [[TMP19:%.*]] = icmp ule <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT2]] 46; IF-EVL-NEXT: [[TMP20:%.*]] = icmp ule <vscale x 4 x i64> [[STEP_ADD]], [[BROADCAST_SPLAT2]] 47; IF-EVL-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i32], ptr [[B:%.*]], <vscale x 4 x i64> [[VEC_IND]], i32 0 48; IF-EVL-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x i32], ptr [[B]], <vscale x 4 x i64> [[STEP_ADD]], i32 0 49; IF-EVL-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> [[TMP21]], i32 4, <vscale x 4 x i1> [[TMP19]], <vscale x 4 x i32> poison) 50; IF-EVL-NEXT: [[WIDE_MASKED_GATHER3:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> [[TMP22]], i32 4, <vscale x 4 x i1> [[TMP20]], <vscale x 4 x i32> poison) 51; IF-EVL-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i32], ptr [[B]], <vscale x 4 x i64> [[VEC_IND]], i32 1 52; IF-EVL-NEXT: [[TMP24:%.*]] = getelementptr inbounds [2 x i32], ptr [[B]], <vscale x 4 x i64> [[STEP_ADD]], i32 1 53; IF-EVL-NEXT: [[WIDE_MASKED_GATHER4:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> [[TMP23]], i32 4, <vscale x 4 x i1> [[TMP19]], <vscale x 4 x i32> poison) 54; IF-EVL-NEXT: [[WIDE_MASKED_GATHER5:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> [[TMP24]], i32 4, <vscale x 4 x i1> [[TMP20]], <vscale x 4 x i32> poison) 55; IF-EVL-NEXT: [[TMP25:%.*]] = add nsw <vscale x 4 x i32> [[WIDE_MASKED_GATHER4]], [[WIDE_MASKED_GATHER]] 56; IF-EVL-NEXT: [[TMP26:%.*]] = add nsw <vscale x 4 x i32> [[WIDE_MASKED_GATHER5]], [[WIDE_MASKED_GATHER3]] 57; IF-EVL-NEXT: [[TMP27:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP13]] 58; IF-EVL-NEXT: [[TMP29:%.*]] = getelementptr inbounds i32, ptr [[TMP27]], i32 0 59; IF-EVL-NEXT: [[TMP30:%.*]] = call i64 @llvm.vscale.i64() 60; IF-EVL-NEXT: [[TMP31:%.*]] = mul i64 [[TMP30]], 4 61; IF-EVL-NEXT: [[TMP32:%.*]] = getelementptr inbounds i32, ptr [[TMP27]], i64 [[TMP31]] 62; IF-EVL-NEXT: call void @llvm.masked.store.nxv4i32.p0(<vscale x 4 x i32> [[TMP25]], ptr [[TMP29]], i32 4, <vscale x 4 x i1> [[TMP19]]) 63; IF-EVL-NEXT: call void @llvm.masked.store.nxv4i32.p0(<vscale x 4 x i32> [[TMP26]], ptr [[TMP32]], i32 4, <vscale x 4 x i1> [[TMP20]]) 64; IF-EVL-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP9]] 65; IF-EVL-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[STEP_ADD]], [[BROADCAST_SPLAT]] 66; IF-EVL-NEXT: [[TMP33:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 67; IF-EVL-NEXT: br i1 [[TMP33]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 68; IF-EVL: middle.block: 69; IF-EVL-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 70; IF-EVL: scalar.ph: 71; IF-EVL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 72; IF-EVL-NEXT: br label [[FOR_BODY:%.*]] 73; IF-EVL: for.body: 74; IF-EVL-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 75; IF-EVL-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], ptr [[B]], i64 [[IV]], i32 0 76; IF-EVL-NEXT: [[TMP34:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 77; IF-EVL-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x i32], ptr [[B]], i64 [[IV]], i32 1 78; IF-EVL-NEXT: [[TMP35:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4 79; IF-EVL-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP35]], [[TMP34]] 80; IF-EVL-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]] 81; IF-EVL-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX4]], align 4 82; IF-EVL-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 83; IF-EVL-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 84; IF-EVL-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 85; IF-EVL: for.cond.cleanup: 86; IF-EVL-NEXT: ret void 87; 88; NO-VP-LABEL: @interleave( 89; NO-VP-NEXT: entry: 90; NO-VP-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 91; NO-VP-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 8 92; NO-VP-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]] 93; NO-VP-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 94; NO-VP: vector.ph: 95; NO-VP-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 96; NO-VP-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 8 97; NO-VP-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]] 98; NO-VP-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 99; NO-VP-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64() 100; NO-VP-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8 101; NO-VP-NEXT: br label [[VECTOR_BODY:%.*]] 102; NO-VP: vector.body: 103; NO-VP-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 104; NO-VP-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 0 105; NO-VP-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64() 106; NO-VP-NEXT: [[TMP8:%.*]] = mul i64 [[TMP7]], 4 107; NO-VP-NEXT: [[TMP9:%.*]] = add i64 [[TMP8]], 0 108; NO-VP-NEXT: [[TMP10:%.*]] = mul i64 [[TMP9]], 1 109; NO-VP-NEXT: [[TMP11:%.*]] = add i64 [[INDEX]], [[TMP10]] 110; NO-VP-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i32], ptr [[B:%.*]], i64 [[TMP6]], i32 0 111; NO-VP-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i32], ptr [[B]], i64 [[TMP11]], i32 0 112; NO-VP-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 8 x i32>, ptr [[TMP12]], align 4 113; NO-VP-NEXT: [[STRIDED_VEC:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC]]) 114; NO-VP-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC]], 0 115; NO-VP-NEXT: [[TMP16:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC]], 1 116; NO-VP-NEXT: [[WIDE_VEC1:%.*]] = load <vscale x 8 x i32>, ptr [[TMP13]], align 4 117; NO-VP-NEXT: [[STRIDED_VEC2:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC1]]) 118; NO-VP-NEXT: [[TMP18:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC2]], 0 119; NO-VP-NEXT: [[TMP19:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC2]], 1 120; NO-VP-NEXT: [[TMP20:%.*]] = add nsw <vscale x 4 x i32> [[TMP16]], [[TMP15]] 121; NO-VP-NEXT: [[TMP21:%.*]] = add nsw <vscale x 4 x i32> [[TMP19]], [[TMP18]] 122; NO-VP-NEXT: [[TMP22:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP6]] 123; NO-VP-NEXT: [[TMP24:%.*]] = getelementptr inbounds i32, ptr [[TMP22]], i32 0 124; NO-VP-NEXT: [[TMP25:%.*]] = call i64 @llvm.vscale.i64() 125; NO-VP-NEXT: [[TMP26:%.*]] = mul i64 [[TMP25]], 4 126; NO-VP-NEXT: [[TMP27:%.*]] = getelementptr inbounds i32, ptr [[TMP22]], i64 [[TMP26]] 127; NO-VP-NEXT: store <vscale x 4 x i32> [[TMP20]], ptr [[TMP24]], align 4 128; NO-VP-NEXT: store <vscale x 4 x i32> [[TMP21]], ptr [[TMP27]], align 4 129; NO-VP-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]] 130; NO-VP-NEXT: [[TMP28:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 131; NO-VP-NEXT: br i1 [[TMP28]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 132; NO-VP: middle.block: 133; NO-VP-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 134; NO-VP-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 135; NO-VP: scalar.ph: 136; NO-VP-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 137; NO-VP-NEXT: br label [[FOR_BODY:%.*]] 138; NO-VP: for.body: 139; NO-VP-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 140; NO-VP-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], ptr [[B]], i64 [[IV]], i32 0 141; NO-VP-NEXT: [[TMP29:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 142; NO-VP-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x i32], ptr [[B]], i64 [[IV]], i32 1 143; NO-VP-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4 144; NO-VP-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP30]], [[TMP29]] 145; NO-VP-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]] 146; NO-VP-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX4]], align 4 147; NO-VP-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 148; NO-VP-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 149; NO-VP-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 150; NO-VP: for.cond.cleanup: 151; NO-VP-NEXT: ret void 152; 153entry: 154 br label %for.body 155 156for.body: 157 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 158 %arrayidx = getelementptr inbounds [2 x i32], ptr %b, i64 %iv, i32 0 159 %0 = load i32, ptr %arrayidx, align 4 160 %arrayidx2 = getelementptr inbounds [2 x i32], ptr %b, i64 %iv, i32 1 161 %1 = load i32, ptr %arrayidx2, align 4 162 %add = add nsw i32 %1, %0 163 %arrayidx4 = getelementptr inbounds i32, ptr %a, i64 %iv 164 store i32 %add, ptr %arrayidx4, align 4 165 %iv.next = add nuw nsw i64 %iv, 1 166 %exitcond.not = icmp eq i64 %iv.next, %N 167 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body, !llvm.loop !0 168 169for.cond.cleanup: 170 ret void 171} 172 173!0 = distinct !{!0, !1, !2} 174!1 = !{!"llvm.loop.interleave.count", i32 2} 175!2 = !{!"llvm.loop.vectorize.enable", i1 true} 176