86 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "^for.body:" --version 32; RUN: opt < %s -mattr=+sve2 -passes=loop-vectorize,instcombine -enable-histogram-loop-vectorization -sve-gather-overhead=2 -sve-scatter-overhead=2 -epilogue-vectorization-minimum-VF=4 -debug-only=loop-vectorize -force-vector-interleave=1 -S 2>&1 | FileCheck %s3; REQUIRES: asserts4 5target triple = "aarch64-unknown-linux-gnu"6 7define void @simple_histogram(ptr noalias %buckets, ptr readonly %indices, i64 %N) {8; CHECK-LABEL: define void @simple_histogram(9; CHECK-SAME: ptr noalias [[BUCKETS:%.*]], ptr readonly [[INDICES:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {10; CHECK-NEXT: iter.check:11; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()12; CHECK-NEXT: [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 113; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]14; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]15; CHECK: vector.main.loop.iter.check:16; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()17; CHECK-NEXT: [[TMP3:%.*]] = shl nuw i64 [[TMP6]], 218; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], [[TMP3]]19; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH1:%.*]]20; CHECK: vector.ph:21; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()22; CHECK-NEXT: [[TMP7:%.*]] = shl nuw i64 [[TMP2]], 223; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP7]]24; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]25; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]26; CHECK: vector.body:27; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]28; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[INDICES]], i64 [[INDEX]]29; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i32>, ptr [[TMP8]], align 430; CHECK-NEXT: [[TMP14:%.*]] = zext <vscale x 4 x i32> [[WIDE_LOAD1]] to <vscale x 4 x i64>31; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], <vscale x 4 x i64> [[TMP14]]32; CHECK-NEXT: call void @llvm.experimental.vector.histogram.add.nxv4p0.i32(<vscale x 4 x ptr> [[TMP15]], i32 1, <vscale x 4 x i1> splat (i1 true))33; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP7]]34; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]35; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]36; CHECK: middle.block:37; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 038; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]39; CHECK: vec.epilog.iter.check:40; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], [[TMP1]]41; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[SCALAR_PH]], label [[VEC_EPILOG_PH]]42; CHECK: vec.epilog.ph:43; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_PH]] ]44; CHECK-NEXT: [[TMP24:%.*]] = call i64 @llvm.vscale.i64()45; CHECK-NEXT: [[TMP25:%.*]] = shl nuw i64 [[TMP24]], 146; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], [[TMP25]]47; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 [[N]], [[N_MOD_VF2]]48; CHECK-NEXT: br label [[FOR_BODY:%.*]]49; CHECK: vec.epilog.vector.body:50; CHECK-NEXT: [[INDEX4:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT6:%.*]], [[FOR_BODY]] ]51; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i32, ptr [[INDICES]], i64 [[INDEX4]]52; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <vscale x 2 x i32>, ptr [[TMP18]], align 453; CHECK-NEXT: [[TMP19:%.*]] = zext <vscale x 2 x i32> [[WIDE_LOAD5]] to <vscale x 2 x i64>54; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], <vscale x 2 x i64> [[TMP19]]55; CHECK-NEXT: call void @llvm.experimental.vector.histogram.add.nxv2p0.i32(<vscale x 2 x ptr> [[TMP20]], i32 1, <vscale x 2 x i1> splat (i1 true))56; CHECK-NEXT: [[INDEX_NEXT6]] = add nuw i64 [[INDEX4]], [[TMP25]]57; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT6]], [[N_VEC3]]58; CHECK-NEXT: br i1 [[TMP21]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]59; CHECK: vec.epilog.middle.block:60; CHECK-NEXT: [[CMP_N7:%.*]] = icmp eq i64 [[N_MOD_VF2]], 061; CHECK-NEXT: br i1 [[CMP_N7]], label [[FOR_EXIT]], label [[SCALAR_PH]]62; CHECK: vec.epilog.scalar.ph:63; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]64; CHECK-NEXT: br label [[FOR_BODY1:%.*]]65; CHECK: for.body:66;67entry:68 br label %for.body69 70for.body:71 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]72 %gep.indices = getelementptr inbounds i32, ptr %indices, i64 %iv73 %l.idx = load i32, ptr %gep.indices, align 474 %idxprom1 = zext i32 %l.idx to i6475 %gep.bucket = getelementptr inbounds i32, ptr %buckets, i64 %idxprom176 %l.bucket = load i32, ptr %gep.bucket, align 477 %inc = add nsw i32 %l.bucket, 178 store i32 %inc, ptr %gep.bucket, align 479 %iv.next = add nuw nsw i64 %iv, 180 %exitcond = icmp eq i64 %iv.next, %N81 br i1 %exitcond, label %for.exit, label %for.body82 83for.exit:84 ret void85}86