Skip to content

/github/workspace/src/FilteringFunctions/kernels/plp_conv_i8p_xpulpv2.c

Functions

Name
void plp_conv_i8p_xpulpv2(void * task_args)
Parallel convolution of 8-bit integer vectors kernel for XPULPV2 extension.

Functions Documentation

function plp_conv_i8p_xpulpv2

void plp_conv_i8p_xpulpv2(
    void * task_args
)

Parallel convolution of 8-bit integer vectors kernel for XPULPV2 extension.

Parameters:

Return: none

Setup code for parallel convolution of 8-bit integer vectors.

Source code

/* =====================================================================
 * Project:      PULP DSP Library
 * Title:        plp_conv_i8p_xpulpv2.c
 * Description:  8-bit parallel integer convolution kernel for XPULPV2
 *
 * $Date:        01. July 2019
 * $Revision:    V0
 *
 * Target Processor: PULP cores
 * ===================================================================== */
/*
 * Copyright (C) 2019 ETH Zurich and University of Bologna.
 *
 * Author: Moritz Scherer, ETH Zurich
 *
 * SPDX-License-Identifier: Apache-2.0
 *
 * Licensed under the Apache License, Version 2.0 (the License); you may
 * not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include "plp_math.h"

// Pre-condition: psrcALen >= psrcBLen, established by calling function plp_conv_i8
// Pre-condition: pRes has enough allocated memory, i.e. srcALen + srcBLen-1u
// Pre-condition: srcALen >= 2 and srcBLen >= 2, otherwise use vector dot product

void plp_conv_i8p_xpulpv2(void *task_args) {

    plp_conv_instance_i8 *S = (plp_conv_instance_i8 *)task_args;

    uint32_t resultoffset = ((S->srcALen + S->nPE - 1) / S->nPE) + S->srcBLen - 1;
    uint32_t srcAoffset = ((S->srcALen + S->nPE - 1) / S->nPE);

    int8_t *pSrcA;
    uint32_t srcALen;
    int8_t *pSrcB;
    uint32_t srcBLen;
    int32_t *pRes;

    int8_t *pIn1;
    int8_t *pIn2;
    uint32_t pIn1Len;
    uint32_t pIn2Len;

    // Unpack partial convolution vectors
    if (hal_core_id() == (S->nPE - 1)) {

        pSrcA = (int8_t *)((S->pSrcA + srcAoffset * (S->nPE - 1)));
        srcALen = S->srcALen - (srcAoffset * (S->nPE - 1));
        pSrcB = (int8_t *)(S->pSrcB);
        srcBLen = S->srcBLen;
        pRes = (int32_t *)(S->pRes + resultoffset * (S->nPE - 1));

    } else {

        srcALen = srcAoffset;
        pSrcA = (int8_t *)(S->pSrcA + (hal_core_id() * srcAoffset));
        pSrcB = (int8_t *)S->pSrcB;
        srcBLen = S->srcBLen;
        pRes = (int32_t *)(S->pRes + resultoffset * (hal_core_id()));
    }

    // Reorder vectors; longest first
    if (srcALen >= srcBLen) {
        pIn1 = pSrcA;
        pIn1Len = srcALen;
        pIn2 = pSrcB;
        pIn2Len = srcBLen;
    } else {
        pIn1 = pSrcB;
        pIn1Len = srcBLen;
        pIn2 = pSrcA;
        pIn2Len = srcALen;
    }

    plp_conv_i8s_xpulpv2(pIn1, pIn1Len, pIn2, pIn2Len, pRes);
    hal_team_barrier();
}

Updated on 2023-03-01 at 16:16:32 +0000