[BUG] Databricks 10.4 aggregations can fail when partially replaced

[jlowe]

While implementing the shim for Databricks 10.4, I encountered test failures when hash aggregates were only partially replaced and contained counts and distinct counts. It appears Databricks 10.4 is using a different intermediate buffer layout and intermediate values are not aligning properly between the GPU and CPU aggregation halves. An example failure (note that I modified test_hash_multiple_mode_query to minimize the repro case with respect to aggregations being used):

_ test_hash_multiple_mode_query[{'spark.rapids.sql.variableFloatAgg.enabled': 'true', 'spark.rapids.sql.hasNans': 'false', 'spark.rapids.sql.castStringToFloat.enabled': 'true', 'spark.rapids.sql.hashAgg.replaceMode': 'partial'}-[('a', Long(not_null)), ('b', Integer(not_null)), ('c', Long(not_null))]] _
[gw1] linux -- Python 3.8.10 /databricks/python3/bin/python

data_gen = [('a', Long(not_null)), ('b', Integer(not_null)), ('c', Long(not_null))]
conf = {'spark.rapids.sql.castStringToFloat.enabled': 'true', 'spark.rapids.sql.hasNans': 'false', 'spark.rapids.sql.hashAgg.replaceMode': 'partial', 'spark.rapids.sql.variableFloatAgg.enabled': 'true'}

    @approximate_float
    @ignore_order
    @incompat
    @pytest.mark.parametrize('data_gen', [_longs_with_no_nulls], ids=idfn)
    @pytest.mark.parametrize('conf', get_params([_no_nans_float_conf_partial], params_markers_for_confs), ids=idfn)
    def test_hash_multiple_mode_query(data_gen, conf):
        print_params(data_gen)
        def doit(spark):
            df = gen_df(spark, data_gen, length=100)\
                .groupby('a')\
                .agg(f.count('a'),
    #                 f.avg('b'),
    #                 f.avg('a'),
                     f.countDistinct('b'),
    #                 f.sum('a'),
    #                 f.min('a'),
    #                 f.max('a'),
    #                 f.sumDistinct('b'),
                     f.countDistinct('c')
                    )
            print("PLAN:", df.explain())
            return df
>       assert_gpu_and_cpu_are_equal_collect(doit, conf=conf)

../../src/main/python/hash_aggregate_test.py:861: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../src/main/python/asserts.py:508: in assert_gpu_and_cpu_are_equal_collect
    _assert_gpu_and_cpu_are_equal(func, 'COLLECT', conf=conf, is_cpu_first=is_cpu_first)
../../src/main/python/asserts.py:439: in _assert_gpu_and_cpu_are_equal
    assert_equal(from_cpu, from_gpu)
../../src/main/python/asserts.py:106: in assert_equal
    _assert_equal(cpu, gpu, float_check=get_float_check(), path=[])
../../src/main/python/asserts.py:42: in _assert_equal
    _assert_equal(cpu[index], gpu[index], float_check, path + [index])
../../src/main/python/asserts.py:35: in _assert_equal
    _assert_equal(cpu[field], gpu[field], float_check, path + [field])
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

cpu = 1, gpu = 0, float_check = <function get_float_check.<locals>.<lambda> at 0x7ffac6c10ca0>, path = [44, 'count(a)']

    def _assert_equal(cpu, gpu, float_check, path):
        t = type(cpu)
        if (t is Row):
            assert len(cpu) == len(gpu), "CPU and GPU row have different lengths at {} CPU: {} GPU: {}".format(path, len(cpu), len(gpu))
            if hasattr(cpu, "__fields__") and hasattr(gpu, "__fields__"):
                assert cpu.__fields__ == gpu.__fields__, "CPU and GPU row have different fields at {} CPU: {} GPU: {}".format(path, cpu.__fields__, gpu.__fields__)
                for field in cpu.__fields__:
                    _assert_equal(cpu[field], gpu[field], float_check, path + [field])
            else:
                for index in range(len(cpu)):
                    _assert_equal(cpu[index], gpu[index], float_check, path + [index])
        elif (t is list):
            assert len(cpu) == len(gpu), "CPU and GPU list have different lengths at {} CPU: {} GPU: {}".format(path, len(cpu), len(gpu))
            for index in range(len(cpu)):
                _assert_equal(cpu[index], gpu[index], float_check, path + [index])
        elif (t is tuple):
            assert len(cpu) == len(gpu), "CPU and GPU list have different lengths at {} CPU: {} GPU: {}".format(path, len(cpu), len(gpu))
            for index in range(len(cpu)):
                _assert_equal(cpu[index], gpu[index], float_check, path + [index])
        elif (t is pytypes.GeneratorType):
            index = 0
            # generator has no zip :( so we have to do this the hard way
            done = False
            while not done:
                sub_cpu = None
                sub_gpu = None
                try:
                    sub_cpu = next(cpu)
                except StopIteration:
                    done = True
    
                try:
                    sub_gpu = next(gpu)
                except StopIteration:
                    done = True
    
                if done:
                    assert sub_cpu == sub_gpu and sub_cpu == None, "CPU and GPU generators have different lengths at {}".format(path)
                else:
                    _assert_equal(sub_cpu, sub_gpu, float_check, path + [index])
    
                index = index + 1
        elif (t is dict):
            # The order of key/values is not guaranteed in python dicts, nor are they guaranteed by Spark
            # so sort the items to do our best with ignoring the order of dicts
            cpu_items = list(cpu.items()).sort(key=_RowCmp)
            gpu_items = list(gpu.items()).sort(key=_RowCmp)
            _assert_equal(cpu_items, gpu_items, float_check, path + ["map"])
        elif (t is int):
>           assert cpu == gpu, "GPU and CPU int values are different at {}".format(path)
E           AssertionError: GPU and CPU int values are different at [44, 'count(a)']

../../src/main/python/asserts.py:77: AssertionError
------------------------------------------------------------------------------ Captured stdout call ------------------------------------------------------------------------------
Test Datagen Params=[('a', 'DataType: LongType, nullable: False, special_cases: [-9223372036854775808, 9223372036854775807, 0, 1, -1]'), ('b', 'DataType: IntegerType, nullable: False, special_cases: [-2147483648, 2147483647, 0, 1, -1]'), ('c', 'DataType: LongType, nullable: False, special_cases: [-9223372036854775808, 9223372036854775807, 0, 1, -1]')]
### CPU RUN ###
== Physical Plan ==
*(3) HashAggregate(keys=[a#0L], functions=[finalmerge_first(merge first#28L, valueSet#29) AS first(count(1)#22L) ignore nulls#23L, finalmerge_count(merge count#31L) AS count(b#20)#13L, finalmerge_count(merge count#33L) AS count(c#21L)#14L])
+- Exchange hashpartitioning(a#0L, 4), ENSURE_REQUIREMENTS, [id=#125]
   +- HashAggregate(keys=[a#0L], functions=[partial_first(count(1)#22L, true) AS (first#28L, valueSet#29) FILTER (WHERE (gid#19 = 0)), partial_count(b#20) AS count#31L FILTER (WHERE (gid#19 = 1)), partial_count(c#21L) AS count#33L FILTER (WHERE (gid#19 = 2))])
      +- *(2) HashAggregate(keys=[a#0L, b#20, c#21L, gid#19], functions=[finalmerge_count(merge count#25L) AS count(1)#9L])
         +- Exchange hashpartitioning(a#0L, b#20, c#21L, gid#19, 4), ENSURE_REQUIREMENTS, [id=#120]
            +- *(1) HashAggregate(keys=[a#0L, b#20, c#21L, gid#19], functions=[partial_count(1) AS count#25L])
               +- *(1) Expand [[a#0L, null, null, 0], [a#0L, b#1, null, 1], [a#0L, null, c#2L, 2]], [a#0L, b#20, c#21L, gid#19]
                  +- *(1) Scan ExistingRDD[a#0L,b#1,c#2L]


PLAN: None
### GPU RUN ###
== Physical Plan ==
*(3) HashAggregate(keys=[a#121L], functions=[finalmerge_first(merge first#149L, valueSet#150) AS first(count(1)#143L) ignore nulls#144L, finalmerge_count(merge count#152L) AS count(b#141)#134L, finalmerge_count(merge count#154L) AS count(c#142L)#135L])
+- GpuColumnarToRow false
   +- GpuShuffleCoalesce 104857600
      +- GpuColumnarExchange gpuhashpartitioning(a#121L, 4), ENSURE_REQUIREMENTS, [id=#520]
         +- GpuHashAggregate(keys=[a#121L], functions=[partial_gpufirst(count(1)#143L, true), partial_gpucount(b#141), partial_gpucount(c#142L)]), filters=ArrayBuffer(Some((gid#140 = 0)), Some((gid#140 = 1)), Some((gid#140 = 2))))
            +- GpuRowToColumnar targetsize(104857600)
               +- *(2) HashAggregate(keys=[a#121L, b#141, c#142L, gid#140], functions=[finalmerge_count(merge count#146L) AS count(1)#130L])
                  +- GpuColumnarToRow false
                     +- GpuShuffleCoalesce 104857600
                        +- GpuColumnarExchange gpuhashpartitioning(a#121L, b#141, c#142L, gid#140, 4), ENSURE_REQUIREMENTS, [id=#512]
                           +- GpuHashAggregate(keys=[a#121L, b#141, c#142L, gid#140], functions=[partial_gpucount(1)]), filters=ArrayBuffer(None))
                              +- GpuExpand [[a#121L, null, null, 0], [a#121L, b#122, null, 1], [a#121L, null, c#123L, 2]], [a#121L, b#141, c#142L, gid#140]
                                 +- GpuRowToColumnar targetsize(104857600)
                                    +- *(1) Scan ExistingRDD[a#121L,b#122,c#123L]


PLAN: None
### COLLECT: GPU TOOK 4.15294623374939 CPU TOOK 10.877075672149658 ###
CPU OUTPUT: [Row(a=-9223372036854775808, count(a)=2, count(b)=2, count(c)=2), Row(a=-8993871743395121636, count(a)=1, count(b)=1, count(c)=1), Row(a=-8824564747028895090, count(a)=1, count(b)=1, count(c)=1), Row(a=-8360439097461740684, count(a)=1, count(b)=1, count(c)=1), Row(a=-8119367677816790003, count(a)=1, count(b)=1, count(c)=1), Row(a=-8095240098983490059, count(a)=1, count(b)=1, count(c)=1), Row(a=-8074900563395293024, count(a)=1, count(b)=1, count(c)=1), Row(a=-7916182562340764963, count(a)=1, count(b)=1, count(c)=1), Row(a=-7644040328848254512, count(a)=1, count(b)=1, count(c)=1), Row(a=-7551557475767929245, count(a)=1, count(b)=1, count(c)=1), Row(a=-7419569242595387455, count(a)=1, count(b)=1, count(c)=1), Row(a=-7130683010620631203, count(a)=1, count(b)=1, count(c)=1), Row(a=-7029864373143088702, count(a)=1, count(b)=1, count(c)=1), Row(a=-6726790004526438438, count(a)=1, count(b)=1, count(c)=1), Row(a=-6681175467107958668, count(a)=1, count(b)=1, count(c)=1), Row(a=-6669292098421320609, count(a)=1, count(b)=1, count(c)=1), Row(a=-6228871387240662005, count(a)=1, count(b)=1, count(c)=1), Row(a=-6150492885085058765, count(a)=1, count(b)=1, count(c)=1), Row(a=-6063156676697350017, count(a)=1, count(b)=1, count(c)=1), Row(a=-5696274501923937029, count(a)=1, count(b)=1, count(c)=1), Row(a=-5545906764439179960, count(a)=1, count(b)=1, count(c)=1), Row(a=-5249581929350276040, count(a)=1, count(b)=1, count(c)=1), Row(a=-5244152136580518692, count(a)=1, count(b)=1, count(c)=1), Row(a=-5147140527860197588, count(a)=1, count(b)=1, count(c)=1), Row(a=-5133656973475552689, count(a)=1, count(b)=1, count(c)=1), Row(a=-4634307270092053473, count(a)=1, count(b)=1, count(c)=1), Row(a=-4594481394522420980, count(a)=1, count(b)=1, count(c)=1), Row(a=-4563575260449999225, count(a)=1, count(b)=1, count(c)=1), Row(a=-4324222002682111580, count(a)=1, count(b)=1, count(c)=1), Row(a=-4320992407945189370, count(a)=1, count(b)=1, count(c)=1), Row(a=-3917032101531217289, count(a)=1, count(b)=1, count(c)=1), Row(a=-3879498789179448709, count(a)=1, count(b)=1, count(c)=1), Row(a=-3854271584414532784, count(a)=1, count(b)=1, count(c)=1), Row(a=-3786594002235441489, count(a)=1, count(b)=1, count(c)=1), Row(a=-3664974252843436884, count(a)=1, count(b)=1, count(c)=1), Row(a=-3502159106106506455, count(a)=1, count(b)=1, count(c)=1), Row(a=-3468876372831096612, count(a)=1, count(b)=1, count(c)=1), Row(a=-3416317927450743577, count(a)=1, count(b)=1, count(c)=1), Row(a=-3210768190877665201, count(a)=1, count(b)=1, count(c)=1), Row(a=-3191579259939627208, count(a)=1, count(b)=1, count(c)=1), Row(a=-2829936275809760174, count(a)=1, count(b)=1, count(c)=1), Row(a=-2828476488387110055, count(a)=1, count(b)=1, count(c)=1), Row(a=-2821373815087770381, count(a)=1, count(b)=1, count(c)=1), Row(a=-2720838936747632101, count(a)=1, count(b)=1, count(c)=1), Row(a=-2602749590540254350, count(a)=1, count(b)=1, count(c)=1), Row(a=-2006832105721144106, count(a)=1, count(b)=1, count(c)=1), Row(a=-1565992008335861162, count(a)=1, count(b)=1, count(c)=1), Row(a=-1522131656554370869, count(a)=1, count(b)=1, count(c)=1), Row(a=-1515104068702799781, count(a)=1, count(b)=1, count(c)=1), Row(a=-948363822265177572, count(a)=1, count(b)=1, count(c)=1), Row(a=-847663416274407371, count(a)=1, count(b)=1, count(c)=1), Row(a=-758486987975055276, count(a)=1, count(b)=1, count(c)=1), Row(a=-545444698343657872, count(a)=1, count(b)=1, count(c)=1), Row(a=-335231079646476073, count(a)=1, count(b)=1, count(c)=1), Row(a=-255208803392541585, count(a)=1, count(b)=1, count(c)=1), Row(a=-199240735297494389, count(a)=1, count(b)=1, count(c)=1), Row(a=-55529886861161874, count(a)=1, count(b)=1, count(c)=1), Row(a=-1, count(a)=1, count(b)=1, count(c)=1), Row(a=1, count(a)=1, count(b)=1, count(c)=1), Row(a=815999380670977075, count(a)=1, count(b)=1, count(c)=1), Row(a=958747012124550583, count(a)=1, count(b)=1, count(c)=1), Row(a=1265497580201624514, count(a)=1, count(b)=1, count(c)=1), Row(a=1362706712718160032, count(a)=1, count(b)=1, count(c)=1), Row(a=1539402556111763558, count(a)=1, count(b)=1, count(c)=1), Row(a=1680065304134987932, count(a)=1, count(b)=1, count(c)=1), Row(a=1819112056103700194, count(a)=1, count(b)=1, count(c)=1), Row(a=2075383307430444458, count(a)=1, count(b)=1, count(c)=1), Row(a=2229783493997241597, count(a)=1, count(b)=1, count(c)=1), Row(a=2251193521986836658, count(a)=1, count(b)=1, count(c)=1), Row(a=2354525851037327742, count(a)=1, count(b)=1, count(c)=1), Row(a=2534571592304619590, count(a)=1, count(b)=1, count(c)=1), Row(a=2889855722692441528, count(a)=1, count(b)=1, count(c)=1), Row(a=3221645619906578280, count(a)=1, count(b)=1, count(c)=1), Row(a=3471682867514641842, count(a)=1, count(b)=1, count(c)=1), Row(a=3852192240925952971, count(a)=1, count(b)=1, count(c)=1), Row(a=4017568188141579524, count(a)=1, count(b)=1, count(c)=1), Row(a=4238750827217987735, count(a)=1, count(b)=1, count(c)=1), Row(a=5051064189658334530, count(a)=1, count(b)=1, count(c)=1), Row(a=5222614686872201741, count(a)=1, count(b)=1, count(c)=1), Row(a=5264124301174436230, count(a)=1, count(b)=1, count(c)=1), Row(a=5331804796221648840, count(a)=1, count(b)=1, count(c)=1), Row(a=5761446890858753124, count(a)=1, count(b)=1, count(c)=1), Row(a=5832380720258166308, count(a)=1, count(b)=1, count(c)=1), Row(a=5904436199519765966, count(a)=1, count(b)=1, count(c)=1), Row(a=6075221653337964625, count(a)=1, count(b)=1, count(c)=1), Row(a=6359360623264328942, count(a)=1, count(b)=1, count(c)=1), Row(a=6365557707853023542, count(a)=1, count(b)=1, count(c)=1), Row(a=6371959128213970013, count(a)=1, count(b)=1, count(c)=1), Row(a=6445989476973674931, count(a)=1, count(b)=1, count(c)=1), Row(a=6446878210185090927, count(a)=1, count(b)=1, count(c)=1), Row(a=7184613054711285292, count(a)=1, count(b)=1, count(c)=1), Row(a=7198729688045931692, count(a)=1, count(b)=1, count(c)=1), Row(a=7258339666698152444, count(a)=1, count(b)=1, count(c)=1), Row(a=7395372434300466361, count(a)=1, count(b)=1, count(c)=1), Row(a=7401983586362591631, count(a)=1, count(b)=1, count(c)=1), Row(a=7618709293599214015, count(a)=1, count(b)=1, count(c)=1), Row(a=7956328026836954420, count(a)=1, count(b)=1, count(c)=1), Row(a=8198116221032872987, count(a)=1, count(b)=1, count(c)=1), Row(a=8886241008042893729, count(a)=1, count(b)=1, count(c)=1)]
GPU OUTPUT: [Row(a=-9223372036854775808, count(a)=2, count(b)=2, count(c)=2), Row(a=-8993871743395121636, count(a)=1, count(b)=1, count(c)=1), Row(a=-8824564747028895090, count(a)=1, count(b)=1, count(c)=1), Row(a=-8360439097461740684, count(a)=1, count(b)=1, count(c)=1), Row(a=-8119367677816790003, count(a)=1, count(b)=1, count(c)=1), Row(a=-8095240098983490059, count(a)=1, count(b)=1, count(c)=1), Row(a=-8074900563395293024, count(a)=1, count(b)=1, count(c)=1), Row(a=-7916182562340764963, count(a)=1, count(b)=1, count(c)=1), Row(a=-7644040328848254512, count(a)=1, count(b)=1, count(c)=1), Row(a=-7551557475767929245, count(a)=1, count(b)=1, count(c)=1), Row(a=-7419569242595387455, count(a)=1, count(b)=1, count(c)=1), Row(a=-7130683010620631203, count(a)=1, count(b)=1, count(c)=1), Row(a=-7029864373143088702, count(a)=1, count(b)=1, count(c)=1), Row(a=-6726790004526438438, count(a)=1, count(b)=1, count(c)=1), Row(a=-6681175467107958668, count(a)=1, count(b)=1, count(c)=1), Row(a=-6669292098421320609, count(a)=1, count(b)=1, count(c)=1), Row(a=-6228871387240662005, count(a)=1, count(b)=1, count(c)=1), Row(a=-6150492885085058765, count(a)=1, count(b)=1, count(c)=1), Row(a=-6063156676697350017, count(a)=1, count(b)=1, count(c)=1), Row(a=-5696274501923937029, count(a)=1, count(b)=1, count(c)=1), Row(a=-5545906764439179960, count(a)=1, count(b)=1, count(c)=1), Row(a=-5249581929350276040, count(a)=1, count(b)=1, count(c)=1), Row(a=-5244152136580518692, count(a)=1, count(b)=1, count(c)=1), Row(a=-5147140527860197588, count(a)=1, count(b)=1, count(c)=1), Row(a=-5133656973475552689, count(a)=1, count(b)=1, count(c)=1), Row(a=-4634307270092053473, count(a)=1, count(b)=1, count(c)=1), Row(a=-4594481394522420980, count(a)=1, count(b)=1, count(c)=1), Row(a=-4563575260449999225, count(a)=1, count(b)=1, count(c)=1), Row(a=-4324222002682111580, count(a)=1, count(b)=1, count(c)=1), Row(a=-4320992407945189370, count(a)=1, count(b)=1, count(c)=1), Row(a=-3917032101531217289, count(a)=1, count(b)=1, count(c)=1), Row(a=-3879498789179448709, count(a)=1, count(b)=1, count(c)=1), Row(a=-3854271584414532784, count(a)=1, count(b)=1, count(c)=1), Row(a=-3786594002235441489, count(a)=1, count(b)=1, count(c)=1), Row(a=-3664974252843436884, count(a)=1, count(b)=1, count(c)=1), Row(a=-3502159106106506455, count(a)=1, count(b)=1, count(c)=1), Row(a=-3468876372831096612, count(a)=1, count(b)=1, count(c)=1), Row(a=-3416317927450743577, count(a)=1, count(b)=1, count(c)=1), Row(a=-3210768190877665201, count(a)=1, count(b)=1, count(c)=1), Row(a=-3191579259939627208, count(a)=1, count(b)=1, count(c)=1), Row(a=-2829936275809760174, count(a)=1, count(b)=1, count(c)=1), Row(a=-2828476488387110055, count(a)=1, count(b)=1, count(c)=1), Row(a=-2821373815087770381, count(a)=1, count(b)=1, count(c)=1), Row(a=-2720838936747632101, count(a)=1, count(b)=1, count(c)=1), Row(a=-2602749590540254350, count(a)=0, count(b)=1, count(c)=1), Row(a=-2006832105721144106, count(a)=1, count(b)=1, count(c)=1), Row(a=-1565992008335861162, count(a)=1, count(b)=1, count(c)=1), Row(a=-1522131656554370869, count(a)=1, count(b)=1, count(c)=1), Row(a=-1515104068702799781, count(a)=1, count(b)=1, count(c)=1), Row(a=-948363822265177572, count(a)=1, count(b)=1, count(c)=1), Row(a=-847663416274407371, count(a)=1, count(b)=1, count(c)=1), Row(a=-758486987975055276, count(a)=1, count(b)=1, count(c)=1), Row(a=-545444698343657872, count(a)=1, count(b)=1, count(c)=1), Row(a=-335231079646476073, count(a)=1, count(b)=1, count(c)=1), Row(a=-255208803392541585, count(a)=1, count(b)=1, count(c)=1), Row(a=-199240735297494389, count(a)=1, count(b)=1, count(c)=1), Row(a=-55529886861161874, count(a)=1, count(b)=1, count(c)=1), Row(a=-1, count(a)=1, count(b)=1, count(c)=1), Row(a=1, count(a)=1, count(b)=1, count(c)=1), Row(a=815999380670977075, count(a)=1, count(b)=1, count(c)=1), Row(a=958747012124550583, count(a)=1, count(b)=1, count(c)=1), Row(a=1265497580201624514, count(a)=1, count(b)=1, count(c)=1), Row(a=1362706712718160032, count(a)=1, count(b)=1, count(c)=1), Row(a=1539402556111763558, count(a)=1, count(b)=1, count(c)=1), Row(a=1680065304134987932, count(a)=1, count(b)=1, count(c)=1), Row(a=1819112056103700194, count(a)=1, count(b)=1, count(c)=1), Row(a=2075383307430444458, count(a)=1, count(b)=1, count(c)=1), Row(a=2229783493997241597, count(a)=1, count(b)=1, count(c)=1), Row(a=2251193521986836658, count(a)=1, count(b)=1, count(c)=1), Row(a=2354525851037327742, count(a)=0, count(b)=1, count(c)=1), Row(a=2534571592304619590, count(a)=1, count(b)=1, count(c)=1), Row(a=2889855722692441528, count(a)=1, count(b)=1, count(c)=1), Row(a=3221645619906578280, count(a)=1, count(b)=1, count(c)=1), Row(a=3471682867514641842, count(a)=1, count(b)=1, count(c)=1), Row(a=3852192240925952971, count(a)=1, count(b)=1, count(c)=1), Row(a=4017568188141579524, count(a)=1, count(b)=1, count(c)=1), Row(a=4238750827217987735, count(a)=1, count(b)=1, count(c)=1), Row(a=5051064189658334530, count(a)=1, count(b)=1, count(c)=1), Row(a=5222614686872201741, count(a)=1, count(b)=1, count(c)=1), Row(a=5264124301174436230, count(a)=1, count(b)=1, count(c)=1), Row(a=5331804796221648840, count(a)=1, count(b)=1, count(c)=1), Row(a=5761446890858753124, count(a)=0, count(b)=1, count(c)=1), Row(a=5832380720258166308, count(a)=1, count(b)=1, count(c)=1), Row(a=5904436199519765966, count(a)=1, count(b)=1, count(c)=1), Row(a=6075221653337964625, count(a)=1, count(b)=1, count(c)=1), Row(a=6359360623264328942, count(a)=1, count(b)=1, count(c)=1), Row(a=6365557707853023542, count(a)=1, count(b)=1, count(c)=1), Row(a=6371959128213970013, count(a)=1, count(b)=1, count(c)=1), Row(a=6445989476973674931, count(a)=1, count(b)=1, count(c)=1), Row(a=6446878210185090927, count(a)=1, count(b)=1, count(c)=1), Row(a=7184613054711285292, count(a)=1, count(b)=1, count(c)=1), Row(a=7198729688045931692, count(a)=1, count(b)=1, count(c)=1), Row(a=7258339666698152444, count(a)=1, count(b)=1, count(c)=1), Row(a=7395372434300466361, count(a)=1, count(b)=1, count(c)=1), Row(a=7401983586362591631, count(a)=1, count(b)=1, count(c)=1), Row(a=7618709293599214015, count(a)=1, count(b)=1, count(c)=1), Row(a=7956328026836954420, count(a)=1, count(b)=1, count(c)=1), Row(a=8198116221032872987, count(a)=1, count(b)=1, count(c)=1), Row(a=8886241008042893729, count(a)=1, count(b)=1, count(c)=1)]

[abellina]

This is a P2 as of now because split aggregates between CPU/GPU were disabled for Databricks 10.4 here: #4974.