Implement vectorized aggregation with grouping on segmentby column (#…

…7147)

This PR implements vectorized aggregation with grouping by segmentby
columns. It works in a limited number of cases: grouping columns also
have to be in the select list, with the order matching the order of
columns in the table, otherwise we run into a projection which prevents
vectorized aggregation.

It also adds `count(*)` function, this is done in the same PR because we
need a second aggregate function to improve testing/code coverage.

.unreleased/group_by_segmentby

@@ -0,0 +1 @@
+Implements: #7147 Vectorize partial aggregation for sum(int4) with grouping on segmentby columns.

src/guc.c

@@ -113,15 +113,16 @@ char *ts_current_timestamp_mock = NULL;
 int ts_guc_debug_toast_tuple_target = 128;
 
 #ifdef TS_DEBUG
-static const struct config_enum_entry require_vector_qual_options[] = {
-	{ "allow", RVQ_Allow, false },
-	{ "forbid", RVQ_Forbid, false },
-	{ "only", RVQ_Only, false },
-	{ NULL, 0, false }
-};
+static const struct config_enum_entry debug_require_options[] = { { "allow", DRO_Allow, false },
+																  { "forbid", DRO_Forbid, false },
+																  { "require", DRO_Require, false },
+																  { NULL, 0, false } };
+
+DebugRequireOption ts_guc_debug_require_vector_qual = DRO_Allow;
+
+DebugRequireOption ts_guc_debug_require_vector_agg = DRO_Allow;
 #endif
 
-DebugRequireVectorQual ts_guc_debug_require_vector_qual = RVQ_Allow;
 bool ts_guc_debug_compression_path_info = false;
 
 static bool ts_guc_enable_hypertable_create = true;
@@ -868,6 +869,19 @@ _guc_init(void)
 							/* assign_hook= */ NULL,
 							/* show_hook= */ NULL);
 
+	DefineCustomEnumVariable(/* name= */ MAKE_EXTOPTION("debug_require_vector_agg"),
+							 /* short_desc= */
+							 "ensure that vectorized aggregation is used or not",
+							 /* long_desc= */ "this is for debugging purposes",
+							 /* valueAddr= */ (int *) &ts_guc_debug_require_vector_agg,
+							 /* bootValue= */ DRO_Allow,
+							 /* options = */ debug_require_options,
+							 /* context= */ PGC_USERSET,
+							 /* flags= */ 0,
+							 /* check_hook= */ NULL,
+							 /* assign_hook= */ NULL,
+							 /* show_hook= */ NULL);
+
 	DefineCustomEnumVariable(/* name= */ MAKE_EXTOPTION("debug_require_vector_qual"),
 							 /* short_desc= */
 							 "ensure that non-vectorized or vectorized filters are used in "
@@ -878,8 +892,8 @@ _guc_init(void)
 							 "and "
 							 "using the test templates is a pain",
 							 /* valueAddr= */ (int *) &ts_guc_debug_require_vector_qual,
-							 /* bootValue= */ RVQ_Allow,
-							 /* options = */ require_vector_qual_options,
+							 /* bootValue= */ DRO_Allow,
+							 /* options = */ debug_require_options,
 							 /* context= */ PGC_USERSET,
 							 /* flags= */ 0,
 							 /* check_hook= */ NULL,

src/guc.h

@@ -76,14 +76,19 @@ extern char *ts_current_timestamp_mock;
 
 extern TSDLLEXPORT int ts_guc_debug_toast_tuple_target;
 
-typedef enum DebugRequireVectorQual
+#ifdef TS_DEBUG
+typedef enum DebugRequireOption
 {
-	RVQ_Allow = 0,
-	RVQ_Forbid,
-	RVQ_Only
-} DebugRequireVectorQual;
+	DRO_Allow = 0,
+	DRO_Forbid,
+	DRO_Require
+} DebugRequireOption;
+
+extern TSDLLEXPORT DebugRequireOption ts_guc_debug_require_vector_qual;
 
-extern TSDLLEXPORT DebugRequireVectorQual ts_guc_debug_require_vector_qual;
+extern TSDLLEXPORT DebugRequireOption ts_guc_debug_require_vector_agg;
+
+#endif
 
 extern TSDLLEXPORT bool ts_guc_debug_compression_path_info;
 

tsl/src/compression/arrow_c_data_interface.h

@@ -136,7 +136,7 @@ struct ArrowSchema
 static pg_attribute_always_inline bool
 arrow_row_is_valid(const uint64 *bitmap, size_t row_number)
 {
-	if (bitmap == NULL)
+	if (likely(bitmap == NULL))
 	{
 		return true;
 	}

tsl/src/nodes/decompress_chunk/planner.c

@@ -1281,11 +1281,11 @@ decompress_chunk_plan_create(PlannerInfo *root, RelOptInfo *rel, CustomPath *pat
 	}
 
 #ifdef TS_DEBUG
-	if (ts_guc_debug_require_vector_qual == RVQ_Forbid && list_length(vectorized_quals) > 0)
+	if (ts_guc_debug_require_vector_qual == DRO_Forbid && list_length(vectorized_quals) > 0)
 	{
 		elog(ERROR, "debug: encountered vector quals when they are disabled");
 	}
-	else if (ts_guc_debug_require_vector_qual == RVQ_Only)
+	else if (ts_guc_debug_require_vector_qual == DRO_Require)
 	{
 		if (list_length(decompress_plan->scan.plan.qual) > 0)
 		{

tsl/src/nodes/decompress_chunk/pred_vector_array.c

@@ -73,7 +73,7 @@ vector_array_predicate(VectorPredicate *vector_const_predicate, bool is_or,
 
 			for (size_t word = 0; word < result_words; word++)
 			{
-				array_result[word] = 0;
+				final_result[word] = 0;
 			}
 			return;
 		}
@@ -95,7 +95,7 @@ vector_array_predicate(VectorPredicate *vector_const_predicate, bool is_or,
 			single_result = single_result_storage;
 			for (size_t outer = 0; outer < result_words; outer++)
 			{
-				single_result[outer] = -1;
+				single_result[outer] = ~0ULL;
 			}
 		}
 		else

tsl/src/nodes/vector_agg/CMakeLists.txt

@@ -1,4 +1,5 @@
 set(SOURCES
     ${CMAKE_CURRENT_SOURCE_DIR}/exec.c ${CMAKE_CURRENT_SOURCE_DIR}/functions.c
+    ${CMAKE_CURRENT_SOURCE_DIR}/grouping_policy_batch.c
     ${CMAKE_CURRENT_SOURCE_DIR}/plan.c)
 target_sources(${TSL_LIBRARY_NAME} PRIVATE ${SOURCES})