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: -force-target-supports-scalable-vectors -scalable-vectorization=on -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 -force-vector-width=4 \ 10; RUN: -force-target-supports-scalable-vectors -scalable-vectorization=on -S < %s | FileCheck --check-prefix=NO-VP %s 11 12; RUN: opt -passes=loop-vectorize \ 13; RUN: -force-tail-folding-style=none \ 14; RUN: -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue \ 15; RUN: -force-target-supports-scalable-vectors -scalable-vectorization=on -S < %s | FileCheck --check-prefix=NO-VP-DEF %s 16 17; The target does not support predicated vectorization. 18define void @foo(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) { 19; IF-EVL-LABEL: @foo( 20; IF-EVL-NEXT: entry: 21; IF-EVL-NEXT: br label [[FOR_BODY:%.*]] 22; IF-EVL: for.body: 23; IF-EVL-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 24; IF-EVL-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[IV]] 25; IF-EVL-NEXT: [[TMP0:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 26; IF-EVL-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[C:%.*]], i64 [[IV]] 27; IF-EVL-NEXT: [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4 28; IF-EVL-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP0]] 29; IF-EVL-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IV]] 30; IF-EVL-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX4]], align 4 31; IF-EVL-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 32; IF-EVL-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N:%.*]] 33; IF-EVL-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY]] 34; IF-EVL: for.cond.cleanup: 35; IF-EVL-NEXT: ret void 36; 37; NO-VP-LABEL: @foo( 38; NO-VP-NEXT: entry: 39; NO-VP-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 40; NO-VP-NEXT: [[TMP8:%.*]] = mul i64 [[TMP0]], 4 41; NO-VP-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP8]] 42; NO-VP-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 43; NO-VP: vector.ph: 44; NO-VP-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64() 45; NO-VP-NEXT: [[TMP14:%.*]] = mul i64 [[TMP1]], 4 46; NO-VP-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP14]] 47; NO-VP-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 48; NO-VP-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 49; NO-VP-NEXT: [[TMP15:%.*]] = mul i64 [[TMP2]], 4 50; NO-VP-NEXT: br label [[VECTOR_BODY:%.*]] 51; NO-VP: vector.body: 52; NO-VP-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 53; NO-VP-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 54; NO-VP-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[TMP3]] 55; NO-VP-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i32 0 56; NO-VP-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP5]], align 4 57; NO-VP-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[C:%.*]], i64 [[TMP3]] 58; NO-VP-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i32 0 59; NO-VP-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i32>, ptr [[TMP7]], align 4 60; NO-VP-NEXT: [[TMP16:%.*]] = add nsw <vscale x 4 x i32> [[WIDE_LOAD1]], [[WIDE_LOAD]] 61; NO-VP-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP3]] 62; NO-VP-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[TMP9]], i32 0 63; NO-VP-NEXT: store <vscale x 4 x i32> [[TMP16]], ptr [[TMP10]], align 4 64; NO-VP-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP15]] 65; NO-VP-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 66; NO-VP-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 67; NO-VP: middle.block: 68; NO-VP-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 69; NO-VP-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 70; NO-VP: scalar.ph: 71; NO-VP-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 72; NO-VP-NEXT: br label [[FOR_BODY:%.*]] 73; NO-VP: for.body: 74; NO-VP-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 75; NO-VP-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV]] 76; NO-VP-NEXT: [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 77; NO-VP-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IV]] 78; NO-VP-NEXT: [[TMP13:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4 79; NO-VP-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP13]], [[TMP12]] 80; NO-VP-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]] 81; NO-VP-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX4]], align 4 82; NO-VP-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 83; NO-VP-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 84; NO-VP-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 85; NO-VP: for.cond.cleanup: 86; NO-VP-NEXT: ret void 87; 88; NO-VP-DEF-LABEL: @foo( 89; NO-VP-DEF-NEXT: entry: 90; NO-VP-DEF-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 91; NO-VP-DEF-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP0]] 92; NO-VP-DEF-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 93; NO-VP-DEF: vector.ph: 94; NO-VP-DEF-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64() 95; NO-VP-DEF-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP1]] 96; NO-VP-DEF-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 97; NO-VP-DEF-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 98; NO-VP-DEF-NEXT: br label [[VECTOR_BODY:%.*]] 99; NO-VP-DEF: vector.body: 100; NO-VP-DEF-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 101; NO-VP-DEF-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 102; NO-VP-DEF-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[TMP3]] 103; NO-VP-DEF-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i32 0 104; NO-VP-DEF-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 1 x i32>, ptr [[TMP5]], align 4 105; NO-VP-DEF-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[C:%.*]], i64 [[TMP3]] 106; NO-VP-DEF-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i32 0 107; NO-VP-DEF-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 1 x i32>, ptr [[TMP7]], align 4 108; NO-VP-DEF-NEXT: [[TMP8:%.*]] = add nsw <vscale x 1 x i32> [[WIDE_LOAD1]], [[WIDE_LOAD]] 109; NO-VP-DEF-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP3]] 110; NO-VP-DEF-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[TMP9]], i32 0 111; NO-VP-DEF-NEXT: store <vscale x 1 x i32> [[TMP8]], ptr [[TMP10]], align 4 112; NO-VP-DEF-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP2]] 113; NO-VP-DEF-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 114; NO-VP-DEF-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 115; NO-VP-DEF: middle.block: 116; NO-VP-DEF-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 117; NO-VP-DEF-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]] 118; NO-VP-DEF: scalar.ph: 119; NO-VP-DEF-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 120; NO-VP-DEF-NEXT: br label [[FOR_BODY:%.*]] 121; NO-VP-DEF: for.body: 122; NO-VP-DEF-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 123; NO-VP-DEF-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV]] 124; NO-VP-DEF-NEXT: [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 125; NO-VP-DEF-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IV]] 126; NO-VP-DEF-NEXT: [[TMP13:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4 127; NO-VP-DEF-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP13]], [[TMP12]] 128; NO-VP-DEF-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]] 129; NO-VP-DEF-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX4]], align 4 130; NO-VP-DEF-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 131; NO-VP-DEF-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 132; NO-VP-DEF-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 133; NO-VP-DEF: for.cond.cleanup: 134; NO-VP-DEF-NEXT: ret void 135; 136entry: 137 br label %for.body 138 139for.body: 140 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 141 %arrayidx = getelementptr inbounds i32, ptr %b, i64 %iv 142 %0 = load i32, ptr %arrayidx, align 4 143 %arrayidx2 = getelementptr inbounds i32, ptr %c, i64 %iv 144 %1 = load i32, ptr %arrayidx2, align 4 145 %add = add nsw i32 %1, %0 146 %arrayidx4 = getelementptr inbounds i32, ptr %a, i64 %iv 147 store i32 %add, ptr %arrayidx4, align 4 148 %iv.next = add nuw nsw i64 %iv, 1 149 %exitcond.not = icmp eq i64 %iv.next, %N 150 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body 151 152for.cond.cleanup: 153 ret void 154} 155