Bug fixes for #14

Author: bwyogatama

src/operator/gpu_physical_grouped_aggregate.cpp

@@ -305,11 +305,14 @@ HandleGroupByAggregateCuDF(vector<shared_ptr<GPUColumn>> &group_by_keys, vector<
 			agg_mode[agg_idx] = AggregationType::SUM;
 		} else if (expr.function.name.compare("sum_no_overflow") == 0 && aggregate_keys[agg_idx]->data_wrapper.data != nullptr) {
 			agg_mode[agg_idx] = AggregationType::SUM;
-			uint64_t* temp = gpuBufferManager->customCudaMalloc<uint64_t>(aggregate_keys[agg_idx]->column_length, 0, 0);
-			convertInt32ToInt64(aggregate_keys[agg_idx]->data_wrapper.data, reinterpret_cast<uint8_t*>(temp), aggregate_keys[agg_idx]->column_length);
-			aggregate_keys[agg_idx]->data_wrapper.data = reinterpret_cast<uint8_t*>(temp);
-			aggregate_keys[agg_idx]->data_wrapper.type = ColumnType::INT64;
-			aggregate_keys[agg_idx]->data_wrapper.num_bytes = aggregate_keys[agg_idx]->data_wrapper.num_bytes * 2;
+			if (aggregate_keys[agg_idx]->data_wrapper.type.id() == GPUColumnTypeId::INT32) {
+				SIRIUS_LOG_DEBUG("Converting INT32 to INT64 for sum_no_overflow");
+				uint64_t* temp = gpuBufferManager->customCudaMalloc<uint64_t>(aggregate_keys[agg_idx]->column_length, 0, 0);
+				convertInt32ToInt64(aggregate_keys[agg_idx]->data_wrapper.data, reinterpret_cast<uint8_t*>(temp), aggregate_keys[agg_idx]->column_length);
+				aggregate_keys[agg_idx]->data_wrapper.data = reinterpret_cast<uint8_t*>(temp);
+				aggregate_keys[agg_idx]->data_wrapper.type = GPUColumnType(GPUColumnTypeId::INT64);
+				aggregate_keys[agg_idx]->data_wrapper.num_bytes = aggregate_keys[agg_idx]->data_wrapper.num_bytes * 2;
+			}
 		} else if (expr.function.name.compare("avg") == 0 && aggregate_keys[agg_idx]->data_wrapper.data != nullptr) {
 			agg_mode[agg_idx] = AggregationType::AVERAGE;
 		} else if (expr.function.name.compare("max") == 0 && aggregate_keys[agg_idx]->data_wrapper.data != nullptr) {

src/operator/gpu_physical_ungrouped_aggregate.cpp

@@ -119,11 +119,14 @@ HandleAggregateExpressionCuDF(vector<shared_ptr<GPUColumn>> &aggregate_keys, GPU
 			agg_mode[agg_idx] = AggregationType::SUM;
 		} else if (expr.function.name.compare("sum_no_overflow") == 0 && aggregate_keys[agg_idx]->data_wrapper.data != nullptr) {
 			agg_mode[agg_idx] = AggregationType::SUM;
-			uint64_t* temp = gpuBufferManager->customCudaMalloc<uint64_t>(aggregate_keys[agg_idx]->column_length, 0, 0);
-			convertInt32ToInt64(aggregate_keys[agg_idx]->data_wrapper.data, reinterpret_cast<uint8_t*>(temp), aggregate_keys[agg_idx]->column_length);
-			aggregate_keys[agg_idx]->data_wrapper.data = reinterpret_cast<uint8_t*>(temp);
-			aggregate_keys[agg_idx]->data_wrapper.type = ColumnType::INT64;
-			aggregate_keys[agg_idx]->data_wrapper.num_bytes = aggregate_keys[agg_idx]->data_wrapper.num_bytes * 2;
+			if (aggregate_keys[agg_idx]->data_wrapper.type.id() == GPUColumnTypeId::INT32) {
+				SIRIUS_LOG_DEBUG("Converting INT32 to INT64 for sum_no_overflow");
+				uint64_t* temp = gpuBufferManager->customCudaMalloc<uint64_t>(aggregate_keys[agg_idx]->column_length, 0, 0);
+				convertInt32ToInt64(aggregate_keys[agg_idx]->data_wrapper.data, reinterpret_cast<uint8_t*>(temp), aggregate_keys[agg_idx]->column_length);
+				aggregate_keys[agg_idx]->data_wrapper.data = reinterpret_cast<uint8_t*>(temp);
+				aggregate_keys[agg_idx]->data_wrapper.type = GPUColumnType(GPUColumnTypeId::INT64);
+				aggregate_keys[agg_idx]->data_wrapper.num_bytes = aggregate_keys[agg_idx]->data_wrapper.num_bytes * 2;
+			}
 		} else if (expr.function.name.compare("avg") == 0 && aggregate_keys[agg_idx]->data_wrapper.data != nullptr) {
 			agg_mode[agg_idx] = AggregationType::AVERAGE;
 		} else if (expr.function.name.compare("max") == 0 && aggregate_keys[agg_idx]->data_wrapper.data != nullptr) {