基于davinci6467的DSP算法移植

只看该作者 · 2017-11-14 14:35

基于davinci6467的DSP算法移植

      工具链是指的Linux下对程序的编译环境。在这里通过 codec_engine_1_02\examples\codecs 下的videnc_copy工程封装DSP.com/forum.php?mod=forumdisplay&fid=58" target="_blank" class="relatedlink">CCS下调试好的程序并生成videnc_copy.a64P，再通过codec_engine_1_02\examples\servers\video_copy工程把videnc_copy.a64P封装成all.x64P
      编译步骤如下：
      1、把ccs下编写好的文件复制到Linux codec_engine_1_02\examples\codecs 下的videnc_copy工程中。
      2.修改该工程中的package.bld文件
      修改var SRCS = ["videnc_copy","test"];括号里只选择工程下的C文件
      修改该文件中的for (var i = 0; i <1 ; i++) { 选择DSP端只用cgtool来编译(在config.bld中C64P排第一个)
      3. 由于输入输出参数由IVIDENC_InArgs，IVIDENC_OutArgs结构体来决定，所以为了增加输出输入参数，应该修改这两个结构体。为此修改xdais_6_25_02_11/packages/ti/xdais/dm中的ividenc.h。
//要添加在结构体的后头

typedef struct IVIDENC_InArgs {
XDAS_Int32 size;
int num1;
int num2;
} IVIDENC_InArgs;
typedef struct IVIDENC_OutArgs {
XDAS_Int32 size;
XDAS_Int32 extendedError;
XDAS_Int32 bytesGenerated;
XDAS_Int32 encodedFrameType;
XDAS_Int32 inputFrameSkip;
IVIDEO_BufDesc reconBufs;
int out;
} IVIDENC_OutArgs;

复制代码

4.在videnc_copy中调用算法：在VIDENCCOPY_TI_process函数中添加(替换原来处理输入输出buf的地方)：

outArgs->out=test(inBufs->bufs[0],outBufs->bufs[0],inBufs->bufSizes[0],inArgs->num1,inArgs->num2);

复制代码

由于使用了#ifdef USE_ACPY3 宏，故而要选择是否使用DMA，为了避免Cache和DMA的冲突，选择不使用DMA。修改

odec_engine_2_25_05_16/examples/ti/sdo/ce/examples/servers/all_codecs下的all.cfg:
VIDENC_COPY.useDMA = false;

复制代码

      5.编译videnc_copy生成videnc_copy.a64P
      6.修改server，编译生成all.x64P
   使用:
      1.修改encode下的Makefile文件，让encode包含server及codec工程:

XDC_PATH = $(USER_XDC_PATH);../../packages;$(DMAI_INSTALL_DIR)/packages;$(CE_INSTALL_DIR)/packages;$(FC_INSTALL_DIR)/packages;$(LINK_INSTALL_DIR)/packages;$(LINK_INSTALL_DIR);$(XDAIS_INSTALL_DIR)/packages;$(CMEM_INSTALL_DIR)/packages;$(CODEC_INSTALL_DIR)/packages;$(CE_INSTALL_DIR)/examples

复制代码

2.修改encode下的encode.cfg:

var demoEngine = Engine.createFromServer(
"test", //引擎句柄
"./bin/ti_platforms_evmDM6467/all.x64P", //以bin开头
"ti.sdo.ce.examples.servers.all_codecs" //路径
);

复制代码

3.文件中调用(例子):

CERuntime_init();
Dmai_init();
VIDENC_Handle hVen = NULL;
Engine_Handle hEngine = NULL;
hEngine = Engine_open("test", NULL, NULL);
if (hEngine == NULL) {
ERR("Failed to open codec engine:test\n");
}
hVen = VIDENC_create(hEngine,"videnc_copy",NULL);
if (hVen == NULL) {
ERR("Failed to create video encoder: videnc_copy");
}
XDM_BufDesc inBufs;
XDM_BufDesc outBufs;
VIDENC_InArgs inArgs;
VIDENC_OutArgs outArgs;
char *JpgBuffer;
char *JpgBuffer1;
JpgBuffer=(char *)Memory_contigAlloc(1024*1024,Memory_DEFAULTALIGNMENT); //在loadmoudle.sh脚本里面要添加相应大小的(arm与dsp)共享内存大小块
JpgBuffer1=(char *)Memory_contigAlloc(1024*1024,Memory_DEFAULTALIGNMENT);
int JpgSize;
FILE *fp=fopen("test.jpg","rb");
fseek(fp,0,SEEK_END);
JpgSize=ftell(fp);
fseek(fp,0,SEEK_SET);
fread(JpgBuffer,1,JpgSize,fp);
fclose(fp);
inBufs.numBufs = 1;
inBufs.bufs = &JpgBuffer;
inBufs.bufSizes = &JpgSize;
outBufs.numBufs = 1;
outBufs.bufs = &JpgBuffer1;
outBufs.bufSizes = &JpgSize;
int num1=1;
int num2=1;
inArgs.num1=num1;
inArgs.num2=num2;
inArgs.size=sizeof(VIDENC_InArgs);
outArgs.size=sizeof(VIDENC_OutArgs);
VIDENC_process (hVen,&inBufs,&outBufs,&inArgs,&outArgs);
if(outArgs.out!=0)
{
printf("copy failed\n");
return -1;
}
FILE *fp1=fopen("out.jpg","wb");
fwrite(JpgBuffer1,1,JpgSize,fp1);
fclose(fp1);

复制代码

附录(算法例子(在Dsp完成简单的图片复制):

int test(char *inbuf,char *outbuf,int size,int num1,int num2)
{
int flag1=num1;
int flag2=num2;
int flag3=-1;
int jpgsize=size;
if(1==flag1&&1==flag2)
{
memcpy(outbuf,inbuf,jpgsize);
flag3=0;
}
return flag3;
}

复制代码

[color=rgb(51, 102, 153) !important]

基于davinci6467的DSP算法移植

相关下载

相关帖子