Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Support GpuCollectList and GpuCollectSet as TypedImperativeAggregate #2971

Merged
merged 17 commits into from
Aug 3, 2021
Merged

Support GpuCollectList and GpuCollectSet as TypedImperativeAggregate #2971

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
e0bdf54
support CollectOps as TypedImperativeAggregate functions
sperlingxx Jul 20, 2021
2cc6ea0
add GpuColumnVector.getRapidsType
sperlingxx Jul 20, 2021
17958ec
add some comments
sperlingxx Jul 20, 2021
6f9c4e4
fix scala code style
sperlingxx Jul 20, 2021
108d122
support associated fallback on TypedImperativeAggregate
sperlingxx Jul 21, 2021
099fc5f
add AQE support
sperlingxx Jul 21, 2021
8a5ae94
Update sql-plugin/src/main/scala/com/nvidia/spark/rapids/GpuOverrides…
sperlingxx Jul 23, 2021
9b469be
Apply suggestions from code review
sperlingxx Jul 23, 2021
246fbf0
Merge remote-tracking branch 'origin/branch-21.10' into gpu_collect_ops
sperlingxx Jul 23, 2021
8c46a93
rework associated fallback
sperlingxx Jul 23, 2021
edadc21
add some comment
sperlingxx Jul 23, 2021
0118cd3
remove ENABLE_TYPED_IMPERATIVE_AGGREGATE
sperlingxx Jul 28, 2021
dadd585
refine
sperlingxx Jul 28, 2021
41a1613
refine
sperlingxx Jul 28, 2021
3bbc9aa
update
sperlingxx Jul 30, 2021
66084d4
test enhencement
sperlingxx Aug 2, 2021
28d3fda
make up some comments
sperlingxx Aug 3, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions docs/configs.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -294,8 +294,8 @@ Name | SQL Function(s) | Description | Default Value | Notes
<a name="sql.expression.Year"></a>spark.rapids.sql.expression.Year|`year`|Returns the year from a date or timestamp|true|None|
<a name="sql.expression.AggregateExpression"></a>spark.rapids.sql.expression.AggregateExpression| |Aggregate expression|true|None|
<a name="sql.expression.Average"></a>spark.rapids.sql.expression.Average|`avg`, `mean`|Average aggregate operator|true|None|
<a name="sql.expression.CollectList"></a>spark.rapids.sql.expression.CollectList|`collect_list`|Collect a list of non-unique elements, only supported in rolling window in current.|true|None|
<a name="sql.expression.CollectSet"></a>spark.rapids.sql.expression.CollectSet|`collect_set`|Collect a set of unique elements, only supported in rolling window in current.|true|None|
<a name="sql.expression.CollectList"></a>spark.rapids.sql.expression.CollectList|`collect_list`|Collect a list of non-unique elements, not supported in reduction.|true|None|
<a name="sql.expression.CollectSet"></a>spark.rapids.sql.expression.CollectSet|`collect_set`|Collect a set of unique elements, not supported in reduction.|true|None|
<a name="sql.expression.Count"></a>spark.rapids.sql.expression.Count|`count`|Count aggregate operator|true|None|
<a name="sql.expression.First"></a>spark.rapids.sql.expression.First|`first_value`, `first`|first aggregate operator|true|None|
<a name="sql.expression.Last"></a>spark.rapids.sql.expression.Last|`last`, `last_value`|last aggregate operator|true|None|
Expand Down Expand Up @@ -327,6 +327,7 @@ Name | Description | Default Value | Notes
<a name="sql.exec.UnionExec"></a>spark.rapids.sql.exec.UnionExec|The backend for the union operator|true|None|
<a name="sql.exec.CustomShuffleReaderExec"></a>spark.rapids.sql.exec.CustomShuffleReaderExec|A wrapper of shuffle query stage|true|None|
<a name="sql.exec.HashAggregateExec"></a>spark.rapids.sql.exec.HashAggregateExec|The backend for hash based aggregations|true|None|
<a name="sql.exec.ObjectHashAggregateExec"></a>spark.rapids.sql.exec.ObjectHashAggregateExec|The backend for hash based aggregations supporting TypedImperativeAggregate functions|true|None|
<a name="sql.exec.SortAggregateExec"></a>spark.rapids.sql.exec.SortAggregateExec|The backend for sort based aggregations|true|None|
<a name="sql.exec.DataWritingCommandExec"></a>spark.rapids.sql.exec.DataWritingCommandExec|Writing data|true|None|
<a name="sql.exec.BatchScanExec"></a>spark.rapids.sql.exec.BatchScanExec|The backend for most file input|true|None|
Expand Down
137 changes: 80 additions & 57 deletions docs/supported_ops.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -495,8 +495,8 @@ Accelerator supports are described below.
<th>UDT</th>
</tr>
<tr>
<td>SortAggregateExec</td>
<td>The backend for sort based aggregations</td>
<td>ObjectHashAggregateExec</td>
<td>The backend for hash based aggregations supporting TypedImperativeAggregate functions</td>
<td>None</td>
<td>S</td>
<td>S</td>
Expand All @@ -512,10 +512,33 @@ Accelerator supports are described below.
<td>S</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><em>PS* (not allowed for grouping expressions; missing nested BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><em>PS* (not allowed for grouping expressions; missing nested BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><em>PS* (not allowed for grouping expressions; missing nested BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><b>NS</b></td>
<td><em>PS (missing nested BOOLEAN, BYTE, SHORT, INT, LONG, FLOAT, DOUBLE, DATE, TIMESTAMP, DECIMAL, NULL, BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
</tr>
<tr>
<td>SortAggregateExec</td>
<td>The backend for sort based aggregations</td>
<td>None</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S*</td>
<td>S</td>
<td>S*</td>
<td>S</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><em>PS* (not allowed for grouping expressions; missing nested BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><em>PS* (not allowed for grouping expressions; missing nested BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><em>PS* (not allowed for grouping expressions; missing nested BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><b>NS</b></td>
</tr>
<tr>
<td>DataWritingCommandExec</td>
Expand Down Expand Up @@ -817,29 +840,6 @@ Accelerator supports are described below.
<td><b>NS</b></td>
</tr>
<tr>
<td>WindowInPandasExec</td>
<td>The backend for Window Aggregation Pandas UDF, Accelerates the data transfer between the Java process and the Python process. It also supports scheduling GPU resources for the Python process when enabled. For now it only supports row based window frame.</td>
<td>This is disabled by default because it only supports row based frame for now</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S*</td>
<td>S</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><em>PS* (missing nested DECIMAL, NULL, BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
</tr>
<tr>
<th>Executor</th>
<th>Description</th>
<th>Notes</th>
Expand All @@ -863,6 +863,29 @@ Accelerator supports are described below.
<th>UDT</th>
</tr>
<tr>
<td>WindowInPandasExec</td>
<td>The backend for Window Aggregation Pandas UDF, Accelerates the data transfer between the Java process and the Python process. It also supports scheduling GPU resources for the Python process when enabled. For now it only supports row based window frame.</td>
<td>This is disabled by default because it only supports row based frame for now</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S*</td>
<td>S</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><em>PS* (missing nested DECIMAL, NULL, BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
</tr>
<tr>
<td>WindowExec</td>
<td>Window-operator backend</td>
<td>None</td>
Expand Down Expand Up @@ -18305,9 +18328,9 @@ Accelerator support is described below.
<tr>
<td rowSpan="6">CollectList</td>
<td rowSpan="6">`collect_list`</td>
<td rowSpan="6">Collect a list of non-unique elements, only supported in rolling window in current.</td>
<td rowSpan="6">Collect a list of non-unique elements, not supported in reduction.</td>
<td rowSpan="6">None</td>
<td rowSpan="2">aggregation</td>
<td rowSpan="2">reduction</td>
<td>input</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
Expand Down Expand Up @@ -18350,25 +18373,25 @@ Accelerator support is described below.
<td> </td>
</tr>
<tr>
<td rowSpan="2">reduction</td>
<td rowSpan="2">aggregation</td>
<td>input</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S*</td>
<td>S</td>
<td>S*</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><em>PS* (missing nested NULL, BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td><b>NS</b></td>
</tr>
<tr>
Expand All @@ -18387,7 +18410,7 @@ Accelerator support is described below.
<td> </td>
<td> </td>
<td> </td>
<td><b>NS</b></td>
<td><em>PS* (missing nested NULL, BINARY, CALENDAR, ARRAY, MAP, UDT)</em></td>
<td> </td>
<td> </td>
<td> </td>
Expand Down Expand Up @@ -18438,9 +18461,9 @@ Accelerator support is described below.
<tr>
<td rowSpan="6">CollectSet</td>
<td rowSpan="6">`collect_set`</td>
<td rowSpan="6">Collect a set of unique elements, only supported in rolling window in current.</td>
<td rowSpan="6">Collect a set of unique elements, not supported in reduction.</td>
<td rowSpan="6">None</td>
<td rowSpan="2">aggregation</td>
<td rowSpan="2">reduction</td>
<td>input</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
Expand Down Expand Up @@ -18483,19 +18506,19 @@ Accelerator support is described below.
<td> </td>
</tr>
<tr>
<td rowSpan="2">reduction</td>
<td rowSpan="2">aggregation</td>
<td>input</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S</td>
<td>S*</td>
<td>S</td>
<td>S*</td>
<td><b>NS</b></td>
<td><b>NS</b></td>
<td><b>NS</b></td>
Expand All @@ -18520,7 +18543,7 @@ Accelerator support is described below.
<td> </td>
<td> </td>
<td> </td>
<td><b>NS</b></td>
<td><em>PS* (missing nested NULL, BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
<td> </td>
<td> </td>
<td> </td>
Expand Down
47 changes: 47 additions & 0 deletions integration_tests/src/main/python/asserts.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -301,6 +301,53 @@ def assert_gpu_fallback_write(write_func,

assert_equal(from_cpu, from_gpu)

def assert_cpu_and_gpu_are_equal_collect_with_capture(func,
exist_classes='',
non_exist_classes='',
conf={}):
(bring_back, collect_type) = _prep_func_for_compare(func, 'COLLECT_WITH_DATAFRAME')

conf = _prep_incompat_conf(conf)

print('### CPU RUN ###')
cpu_start = time.time()
from_cpu, cpu_df = with_cpu_session(bring_back, conf=conf)
cpu_end = time.time()
print('### GPU RUN ###')
gpu_start = time.time()
from_gpu, gpu_df = with_gpu_session(bring_back, conf=conf)
gpu_end = time.time()
jvm = spark_jvm()
for clz in exist_classes.split(','):
jvm.com.nvidia.spark.rapids.ExecutionPlanCaptureCallback.assertContains(gpu_df._jdf, clz)
for clz in non_exist_classes.split(','):
jvm.com.nvidia.spark.rapids.ExecutionPlanCaptureCallback.assertNotContain(gpu_df._jdf, clz)
print('### {}: GPU TOOK {} CPU TOOK {} ###'.format(collect_type,
gpu_end - gpu_start, cpu_end - cpu_start))
if should_sort_locally():
from_cpu.sort(key=_RowCmp)
from_gpu.sort(key=_RowCmp)

assert_equal(from_cpu, from_gpu)

def assert_cpu_and_gpu_are_equal_sql_with_capture(df_fun,
sql,
table_name,
exist_classes='',
non_exist_classes='',
conf=None,
debug=False):
if conf is None:
conf = {}
def do_it_all(spark):
df = df_fun(spark)
df.createOrReplaceTempView(table_name)
if debug:
return data_gen.debug_df(spark.sql(sql))
else:
return spark.sql(sql)
assert_cpu_and_gpu_are_equal_collect_with_capture(do_it_all, exist_classes, non_exist_classes, conf)

def assert_gpu_fallback_collect(func,
cpu_fallback_class_name,
conf={}):
Expand Down
Loading