kupl_parallel_for_reduce

Perform parallel reduction.

Interface Definition

int kupl_parallel_for_reduce(kupl_parallel_for_desc_t *desc, kupl_pf_reduce_func_t func, void *args,

kupl_reduce_args_t *rd_args);

Parameters

**Table 1** Parameter definition
Parameter	Type	Description	Input/Output
desc	kupl_parallel_for_desc_t *	Description of a parallel iteration task, which is a pointer to the kupl_parallel_for_desc_t structure. For details, see the definition in the section "kupl_parallel_for."	Input
func	kupl_pf_reduce_func_t	Parallel reduction function, which must be defined in the following format of "void (kupl_pf_reduce_func_t)(kupl_nd_range_t nd_range, void args, int tid, int tnum, kupl_reduce_args_t rd_args)". nd_range indicates the loop region actually executed by the function after internal processing via kupl_parallel_for, args indicates the arguments passed to this structure, tid and tnum indicate the ID of the thread where the current function is executed and the total number of threads, respectively, and rd_args indicates the thread replica of the reduction parameter. For details, see the definition of kupl_reduce_args_t.	Input
args	void *	Arguments to be passed to func.	Input
rd_args	kupl_reduce_args_t *	Reduction parameter, which defines the total number of reduction operations, data type of the reduction operation, reduction operation type, and reduction data address.	Input/Output

**Table 2** kupl_reduce_args_t structure definition
Parameter	Type	Description
num	int	Number of reduction operations.
items	kupl_reduce_item_t *	Pointer to the specific reduction operation definition. For details, see the following description.

**Table 3** kupl_reduce_item_t structure definition
Parameter	Type	Description
buffer	void *	Data address of the reduction operation.
type	kupl_datatype_t	Data type of the reduction operation, including: KUPL_DATATYPE_INT, KUPL_DATATYPE_FLOAT, KUPL_DATATYPE_DOUBLE, KUPL_DATATYPE_FLOAT_COMPLEX, KUPL_DATATYPE_DOUBLE_COMPLEX
op	kupl_reduce_op_t	Reduction operation type, including: KUPL_RD_ADD, KUPL_RD_SUB, KUPL_RD_MAX, KUPL_RD_MIN

Return Value

Success: KUPL_OK
Failure: KUPL_ERROR

Examples

One-dimensional loop:

#include <stdio.h>
#include <pthread.h>
#include "kupl.h"
static inline void task_in_loop_reduce(kupl_nd_range_t *nd_range, void *args, int tid, int tnum, kupl_reduce_args_t *rd_args)
{
    int *data = (int *)args;
    int localsum = 0;
    for (int i = nd_range->nd_range[0].lower; i < nd_range->nd_range[0].upper; i += nd_range->nd_range[0].step) {
        localsum += data[i];
    }
    *(int *)rd_args->items[0].buffer += localsum;
}

int main()
{
    const int n = 100;
    int data[n];
    for (int i = 0; i < n; i++) {
        data[i] = i + 1;
    }
    kupl_nd_range_t range;
    KUPL_1D_RANGE_INIT(range, 0, n);


    kupl_parallel_for_desc_t desc = {
        .field_mask = KUPL_PARALLEL_FOR_DESC_FIELD_DEFAULT,
        .range = &range,
        .egroup = NULL,
        .concurrency = 4,
        .policy = KUPL_LOOP_POLICY_STATIC
    };
    int sum_int = 0;
    kupl_reduce_item_t param_int[1] = {{ .buffer = &sum_int, .type = KUPL_DATATYPE_INT, .op = KUPL_RD_ADD }};
    kupl_reduce_args_t rd_args_int = { .num = 1, .items = param_int };
    kupl_parallel_for_reduce(&desc, task_in_loop_reduce, data, &rd_args_int);
    printf("sum: %d\n", sum_int);
}

The execution result is as follows:

sum: 5050

The preceding example demonstrates the process of KUPL parallel reduction and calculates the sum of all values from 1 to 100 in the data array.

Parent topic: Multi-threaded Programming Functions