87 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "^for.body:"2; RUN: opt < %s -passes=loop-vectorize -mtriple aarch64-unknown-linux-gnu -mattr=+sve -epilogue-vectorization-force-VF=2 -S | FileCheck %s3 4;5; Strict fadd reduction with interleaving6;7define float @fadd_strict(ptr noalias nocapture readonly %a, i64 %n) {8; CHECK-LABEL: @fadd_strict(9; CHECK-NEXT: iter.check:10; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 211; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]12; CHECK: vector.main.loop.iter.check:13; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()14; CHECK-NEXT: [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 315; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], [[TMP1]]16; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]17; CHECK: vector.ph:18; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()19; CHECK-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 820; 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: [[VEC_PHI:%.*]] = phi float [ 0xFFFFFFFFE0000000, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]26; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[INDEX]]27; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()28; CHECK-NEXT: [[TMP16:%.*]] = shl nuw i64 [[TMP15]], 229; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[TMP12]], i64 [[TMP16]]30; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x float>, ptr [[TMP12]], align 431; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x float>, ptr [[TMP17]], align 432; CHECK-NEXT: [[TMP18:%.*]] = call float @llvm.vector.reduce.fadd.nxv4f32(float [[VEC_PHI]], <vscale x 4 x float> [[WIDE_LOAD]])33; CHECK-NEXT: [[TMP19]] = call float @llvm.vector.reduce.fadd.nxv4f32(float [[TMP18]], <vscale x 4 x float> [[WIDE_LOAD2]])34; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]35; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]36; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]37; CHECK: middle.block:38; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]39; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]40; CHECK: vec.epilog.iter.check:41; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 242; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]43; CHECK: vec.epilog.ph:44; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]45; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP19]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0xFFFFFFFFE0000000, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]46; CHECK-NEXT: [[N_MOD_VF3:%.*]] = urem i64 [[N]], 247; CHECK-NEXT: [[N_VEC4:%.*]] = sub i64 [[N]], [[N_MOD_VF3]]48; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]49; CHECK: vec.epilog.vector.body:50; CHECK-NEXT: [[INDEX6:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT9:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]51; CHECK-NEXT: [[VEC_PHI7:%.*]] = phi float [ [[BC_MERGE_RDX]], [[VEC_EPILOG_PH]] ], [ [[TMP24:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]52; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX6]]53; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x float>, ptr [[TMP22]], align 454; CHECK-NEXT: [[TMP24]] = call float @llvm.vector.reduce.fadd.v2f32(float [[VEC_PHI7]], <2 x float> [[WIDE_LOAD8]])55; CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[INDEX6]], 256; CHECK-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT9]], [[N_VEC4]]57; CHECK-NEXT: br i1 [[TMP25]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]58; CHECK: vec.epilog.middle.block:59; CHECK-NEXT: [[CMP_N5:%.*]] = icmp eq i64 [[N]], [[N_VEC4]]60; CHECK-NEXT: br i1 [[CMP_N5]], label [[FOR_END]], label [[VEC_EPILOG_SCALAR_PH]]61; CHECK: vec.epilog.scalar.ph:62; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC4]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]63; CHECK-NEXT: [[BC_MERGE_RDX10:%.*]] = phi float [ [[TMP24]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[TMP19]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0xFFFFFFFFE0000000, [[ITER_CHECK]] ]64; CHECK-NEXT: br label [[FOR_BODY:%.*]]65; CHECK: for.body:66;67entry:68 br label %for.body69 70for.body:71 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]72 %sum.07 = phi float [ 0xFFFFFFFFE0000000, %entry ], [ %add, %for.body ]73 %arrayidx = getelementptr inbounds float, ptr %a, i64 %iv74 %0 = load float, ptr %arrayidx, align 475 %add = fadd float %0, %sum.0776 %iv.next = add nuw nsw i64 %iv, 177 %exitcond.not = icmp eq i64 %iv.next, %n78 br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !079 80for.end:81 ret float %add82}83 84!0 = distinct !{!0, !1, !2}85!1 = !{!"llvm.loop.interleave.count", i32 2}86!2 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}87