[GLUTEN-4350][CH] Improve NativeReader's performance for small blocks

[lgbo-ustc]

What changes were proposed in this pull request?

(Please fill in changes proposed in this fix)

Fixes: #4350

After this modification.

How was this patch tested?

(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)

unit tests

(If this patch involves UI changes, please attach a screenshot; otherwise, remove this)

[github-actions]

#4350

[github-actions]

Run Gluten Clickhouse CI

[github-actions]

Run Gluten Clickhouse CI

[github-actions]

Run Gluten Clickhouse CI

[lgbo-ustc]

Notice that when parse aggregate data, there are a lot of virtual function call to initialize an aggregate data place. Reduce these virtual function call should bring some improvement.

Run a test on this.

static void readFixedSizeAggregateData(DB::ReadBuffer &in, DB::ColumnPtr & column, size_t rows, NativeReader::ColumnParseUtil & column_parse_util)
{
    ColumnAggregateFunction & real_column = typeid_cast<ColumnAggregateFunction &>(*column->assumeMutable());
    auto & arena = real_column.createOrGetArena();
    ColumnAggregateFunction::Container & vec = real_column.getData();
    size_t initial_size = vec.size();
    vec.reserve(initial_size + rows);
    #if 0
    for (size_t i = 0; i < rows; ++i)
    {
        AggregateDataPtr place = arena.alignedAlloc(column_parse_util.aggregate_state_size, column_parse_util.aggregate_state_align);
        column_parse_util.aggregate_function->create(place);
        auto n = in.read(place, column_parse_util.aggregate_state_size);
        chassert(n == column_parse_util.aggregate_state_size);
        vec.push_back(place);
    }
    #else
    AggregateDataPtr places = arena.alignedAlloc(column_parse_util.aggregate_state_align_size * rows, column_parse_util.aggregate_state_align);
    column_parse_util.aggregate_function->createPlaces(places, rows, column_parse_util.aggregate_state_align_size);
    for (size_t i = 0; i < rows; ++i)
    {
        auto n = in.read(places, column_parse_util.aggregate_state_size);
        chassert(n == column_parse_util.aggregate_state_size);
        places += column_parse_util.aggregate_state_align_size;
        vec.push_back(places);
    }
    #endif
}

                if (!column_parse_util.aggregate_state_align)
                    column_parse_util.aggregate_state_align_size = column_parse_util.aggregate_state_size;
                else
                {
                    auto n = column_parse_util.aggregate_state_size % column_parse_util.aggregate_state_align;
                    if (!n)
                    {
                        column_parse_util.aggregate_state_align_size = column_parse_util.aggregate_state_size + (column_parse_util.aggregate_state_align - n);
                    }
                    else
                        column_parse_util.aggregate_state_align_size = column_parse_util.aggregate_state_size;
                }

Got result

But need to modify CH core.

[github-actions]

Run Gluten Clickhouse CI

[github-actions]

Run Gluten Clickhouse CI

[baibaichen]

LGTM