search

rockchip-linux / mpp

Media Process Platform (MPP) module
526 stars 161 forks source link

用mpi_dec_test解码jpg图片失败 #79

Open niujiabenben opened 5 years ago

niujiabenben commented 5 years ago

我想用mpp解码单个jpg图片, 操作过程为: git clone -b release https://github.com/rockchip-linux/mpp.git cmake -DRKPLATFORM=ON -DHAVE_DRM=ON && make ./test/mpi_dec_test -i test.jpg -w 1280 -h 720 -t 8

报错: mpi_dec_test: cmd parse result: mpi_dec_test: input file name: test.jpg mpi_dec_test: output file name: mpi_dec_test: config file name: mpi_dec_test: width : 1280 mpi_dec_test: height : 720 mpi_dec_test: type : 8 mpi_dec_test: debug flag : 0 mpi_dec_test: max frames : 0 mpi_dec_test: mpi_dec_test start mpi_dec_test: input file size 290872 mpp_rt: NOT found ion allocator mpp_rt: found drm allocator mpi_dec_test: mpi_dec_test decoder test start w 1280 h 720 type 8 mpi: mpp version: 10c9a2d author: Herman Chen [jpegd]: Add vpu jpegd stream patch mpp_log: can not found match soc name: rockchip,android 9tripod,x3399-development-board rockchip,rk339 mpp_dec: mpp_dec_advanced_thread slot size 1843200 is not equal to buffer size 3686400 mpp_dec: Assertion slot_size == buffer_size failed at mpp_dec_advanced_thread:1086 然后程序就卡在这里不动了....

我所用的系统: Linux X3399 4.4.83 #7 SMP PREEMPT Wed Nov 21 17:49:40 CST 2018 aarch64 aarch64 aarch64 GNU/Linux

请问该如何解决? 谢谢.

HermanChen commented 5 years ago

麻烦提供一下测试源

niujiabenben commented 5 years ago

测试图片 (下载解压即可): test.zip

解压后图片的md5: cdd6ccc6e346a1ced102a42ad24fee4b

HermanChen commented 5 years ago

在命令行指定下 -f 4 的输出格式可以正常，具体问题还在处理中。

HermanChen commented 5 years ago

tmp.zip 试一下这个补丁，如果还有问题的话提供下log，可能是内核相关的问题

HermanChen commented 5 years ago

加了 patch 之后，在新版本内核上解码正常，估计是内核版本问题

eddiecailinux commented 5 years ago

麻烦更新最新的kernel测试

niujiabenben commented 5 years ago

目前我们用的板子是定制的, 更新kernel有一些麻烦. 在当前的系统下, 解码jpeg图片还是有问题.

我往前找了几个commit, 比如 9月5号的这个commit: 15e357ffd0c98e3099be81eb91b4a48cf8db78a7 没有上述的 "Assertion slot_size == buffer_size failed " 问题了, 但是程序卡在下面的语句出不去了: ret = mpi->poll(ctx, MPP_PORT_OUTPUT, MPP_POLL_BLOCK);

请问您能提供一个简单的在我们当前系统下能运行的demo么? 或者给一些建议我试着自己实现以下? 十分感谢!

HermanChen commented 5 years ago

https://github.com/HermanChen/mpp/ 用我自己个人的开发版本看一下有没有问题，加了一个 [hal_jpegd]: Improve address setup 的提交 在 3288 平台上测试正常

niujiabenben commented 5 years ago

还是不行, 和上面一样卡在这一句出不去了: ret = mpi->poll(ctx, MPP_PORT_OUTPUT, MPP_POLL_BLOCK);

HermanChen commented 5 years ago

我在手上的 3228 平台（和 3399 一样的 vpu硬件）上做测试，会出现解码花的情况，但是不会卡死，现象有不同啊

HermanChen commented 5 years ago

export jpegd_debug=255，然后再跑一下看看

niujiabenben commented 5 years ago

版本: https://github.com/HermanChen/mpp/ , 加上export jpegd_debug=255之后, 输出: mpi_dec_test: cmd parse result: mpi_dec_test: input file name: test.jpg mpi_dec_test: output file name: mpi_dec_test: config file name: mpi_dec_test: width : 1280 mpi_dec_test: height : 720 mpi_dec_test: type : 8 mpi_dec_test: debug flag : 0 mpi_dec_test: max frames : 0 mpi_dec_test: mpi_dec_test start mpi_dec_test: input file size 290872 mpp_rt: NOT found ion allocator mpp_rt: found drm allocator mpi_dec_test: mpi_dec_test decoder test start w 1280 h 720 type 8 mpi: mpp version: b03dd17 author: Herman Chen [hal_jpegd]: Improve address setup mpp_log: can not found match soc name: rockchip,android 9tripod,x3399-development-board rockchip,rk339 jpegd_parser: jpegd_init exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_init enter HAL_JPEGD_VDPU1: jpegd_regs_init enter HAL_JPEGD_VDPU1: jpegd_regs_init exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_init exit jpegd_parser: jpegd_prepare enter jpegd_parser: pkt_length 290872 eos 0 jpegd_parser: jpegd_split_frame enter jpegd_parser: jpegd_split_frame exit jpegd_parser: input_packet:0x42efa0, recv_buffer:0x7f9533d040, pkt_length:290872 jpegd_parser: jpegd_prepare exit jpegd_parser: jpegd_parse enter jpegd_parser: jpegd_decode_frame enter jpegd_parser: find_marker skipped 0 bytes jpegd_parser: marker = 0xd8, avail_size_in_buf = 290870 jpegd_parser: find_marker skipped 0 bytes jpegd_parser: marker = 0xe0, avail_size_in_buf = 290868 jpegd_parser: mjpeg: JFIF header found (version: 1.1) SAR=1/1 jpegd_parser: unhandled coding type(0xe0) in switch..case.. jpegd_parser: find_marker skipped 16 bytes jpegd_parser: marker = 0xdb, avail_size_in_buf = 290850 jpegd_parser: quantize tables ID=0 jpegd_parser: **Start to print quantize table 0**

0x02,0x01,0x01,0x01,0x01,0x01,0x02,0x01, 0x01,0x01,0x02,0x02,0x02,0x02,0x02,0x04, 0x03,0x02,0x02,0x02,0x02,0x05,0x04,0x04, 0x03,0x04,0x06,0x05,0x06,0x06,0x06,0x05, 0x06,0x06,0x06,0x07,0x09,0x08,0x06,0x07, 0x09,0x07,0x06,0x06,0x08,0x0b,0x08,0x09, 0x0a,0x0a,0x0a,0x0a,0x0a,0x06,0x08,0x0b, 0x0c,0x0b,0x0a,0x0c,0x09,0x0a,0x0a,0x0a,jpegd_parser: **Quantize table 0 End**

jpegd_parser: qscale[0]: 0 jpegd_parser: find_marker skipped 67 bytes jpegd_parser: marker = 0xdb, avail_size_in_buf = 290781 jpegd_parser: quantize tables ID=1 jpegd_parser: **Start to print quantize table 1**

0x02,0x02,0x02,0x02,0x02,0x02,0x05,0x03, 0x03,0x05,0x0a,0x07,0x06,0x07,0x0a,0x0a, 0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a, 0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a, 0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a, 0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a, 0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a, 0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,0x0a,jpegd_parser: **Quantize table 1 End**

jpegd_parser: qscale[1]: 1 jpegd_parser: find_marker skipped 67 bytes jpegd_parser: marker = 0xc0, avail_size_in_buf = 290712 jpegd_parser: sof0: picture: 1280x720, stride: 1280x720 jpegd_parser: component 0 2:2 id: 0 quant_id:0 jpegd_parser: component 1 1:1 id: 1 quant_id:1 jpegd_parser: component 2 1:1 id: 2 quant_id:1 jpegd_parser: YCbCr Format: YUV420(22:11:1*1) jpegd_parser: find_marker skipped 17 bytes jpegd_parser: marker = 0xc4, avail_size_in_buf = 290693 jpegd_parser: dht: huffman tables length=29 jpegd_parser: dht: type=0 id=0 code_word_num=12, code_max=11, len=0 jpegd_parser: find_marker skipped 31 bytes jpegd_parser: marker = 0xc4, avail_size_in_buf = 290660 jpegd_parser: dht: huffman tables length=179 jpegd_parser: dht: type=1 id=0 code_word_num=162, code_max=250, len=0 jpegd_parser: find_marker skipped 181 bytes jpegd_parser: marker = 0xc4, avail_size_in_buf = 290477 jpegd_parser: dht: huffman tables length=29 jpegd_parser: dht: type=0 id=1 code_word_num=12, code_max=11, len=0 jpegd_parser: find_marker skipped 31 bytes jpegd_parser: marker = 0xc4, avail_size_in_buf = 290444 jpegd_parser: dht: huffman tables length=179 jpegd_parser: dht: type=1 id=1 code_word_num=162, code_max=250, len=0 jpegd_parser: find_marker skipped 181 bytes jpegd_parser: marker = 0xda, avail_size_in_buf = 290261 jpegd_parser: sos component: 0 jpegd_parser: component:0, dc_index:0, ac_index:0 jpegd_parser: sos component: 1 jpegd_parser: component:1, dc_index:1, ac_index:1 jpegd_parser: sos component: 2 jpegd_parser: component:2, dc_index:1, ac_index:1 jpegd_parser: This packet owns 290872 bytes, has been decoded 623 bytes by software buf_ptr:0x7f9533d2a3, buf:0x7f9533d040, sos_len:12 hardware start address:0x7f9533d2af jpegd_parser: find_marker skipped 290259 bytes jpegd_parser: marker = 0xd9, avail_size_in_buf = 0 jpegd_parser: still exists 0 bytes behind EOI marker jpegd_parser: jpegd_decode_frame exit jpegd_parser: jpegd_allocate_frame enter jpegd_parser: frame_slot_index:0 jpegd_parser: jpegd_allocate_frame exit jpegd_parser: jpegd_update_frame enter jpegd_parser: jpegd_update_frame exit jpegd_parser: jpegd_parse exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_gen_regs enter HAL_JPEGD_COMMON: jpegd_setup_output_fmt enter HAL_JPEGD_COMMON: jpegd_setup_output_fmt exit HAL_JPEGD_VDPU1: jpegd_setup_pp enter HAL_JPEGD_VDPU1: output_frame_fd:7, reg14:38400007 HAL_JPEGD_VDPU1: jpegd_setup_pp exit HAL_JPEGD_VDPU1: jpegd_gen_regs enter HAL_JPEGD_VDPU1: jpegd_write_code_word_number enter HAL_JPEGD_VDPU1: jpegd_write_code_word_number exit HAL_JPEGD_COMMON: jpegd_write_qp_ac_dc_table enter HAL_JPEGD_COMMON: jpegd_write_qp_ac_dc_table exit HAL_JPEGD_VDPU1: jpegd_set_chroma_table_id enter HAL_JPEGD_VDPU1: jpegd_set_chroma_table_id exit HAL_JPEGD_VDPU1: jpegd_set_stream_offset enter HAL_JPEGD_VDPU1: jpegd_set_stream_offset exit HAL_JPEGD_VDPU1: jpegd_gen_regs exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_gen_regs exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_start enter

niujiabenben commented 5 years ago

如果您愿意的话, 我们可以寄一块板子过去给您测试.

HermanChen commented 5 years ago

产品支持最好走公司的 redmine 来提交问题单，在 github 不合适做正规的客户支持

alandingbc commented 5 years ago

mpp解码jpeg图像，结果提示可能是png图像。请问这个是什么问题？

:~/mpp$ ./test/mpi_dec_test -i ./1080.jpg -t 8 -w 1920 -h 1080 mpi_dec_test: cmd parse result: mpi_dec_test: input file name: ./1080.jpg mpi_dec_test: output file name: mpi_dec_test: config file name: mpi_dec_test: width : 1920 mpi_dec_test: height : 1080 mpi_dec_test: type : 8 mpi_dec_test: debug flag : 0 mpi_dec_test: max frames : 0 mpi_dec_test: mpi_dec_test start mpi_dec_test: input file size 340928 mpp_rt: NOT found ion allocator mpp_rt: found drm allocator mpi_dec_test: mpi_dec_test decoder test start w 1920 h 1080 type 8 mpi: mpp version: 930a2737 author: Ding Wei [h264d]: fix bug: fast mode buffer malloc jpegd_parser: input img maybe png format,check it jpegd_parser: input img maybe png format,check it mpp_dec: mpp_dec_advanced_thread slot size 4177920 is not equal to buffer size 8355840 mpp_dec: Assertion slot_size == buffer_size failed at mpp_dec_advanced_thread:1086 mpi_dec_test: decoded frame 1 mpi_dec_test: found last packet mpi_dec_test: decoded frame 2 mpi_dec_test: found eos frame mpi_dec_test: test success max memory 0.00 MB

HermanChen commented 5 years ago

只是 warning，没事，有客户把 png 当 jpeg 送进来过，于是加了这个打印

alandingbc commented 5 years ago

但是，后面size大小 assert信息，包括 test success max memory 0.00 MB也是正常的嘛？

HermanChen commented 5 years ago

正常，jpeg 解码使用的是 mpp 外部提供的 buffer，mpp 内部本身不分配新 buffer，所以使用量为 0

alandingbc commented 5 years ago

额、我再结文档看看代码，mpp buffer的三种管理模式不是清楚。谢谢

HermanChen commented 5 years ago

MPP 开发参考_v0.3.pdf 文档供参考

yuanouwen commented 5 years ago

我看mpp_dec_test.cpp测试对MJPEG的解码单独做了处理，这个原因是啥，还有我用此测试程序解码MJPEG,发现耗时达60ms，720p像素，请解答下

yuanouwen commented 5 years ago

我看mpp_dec_test.cpp测试对MJPEG的解码单独做了处理，这个原因是啥，还有我用此测试程序解码MJPEG,发现耗时达60ms，720p像素，请解答下

HermanChen commented 5 years ago

mjpeg 解码是单次解码，buffer 是外部应用提供的，与一般的解码器不同 你看到的时间估计包括了码流拷贝，buffer 分配等的软件时间，硬件时间会比较短，多次解码情况下，流水处理时间会短

yuanouwen commented 5 years ago

发现 ####mpi_dec time7 将近40ms。####sencond memcpy time 为20ms。所以导致硬解差不多60ms。请帮忙分析下。代码如下： static int decode_advanced(MpiDecLoopData data,void imageData,int imageLen,void returnData,int returnLen) { RK_U32 pkt_eos = 0; MPP_RET ret = MPP_OK; MppCtx ctx = data->ctx; MppApi mpi = data->mpi; char buf = data->buf; MppPacket packet = data->packet; MppFrame frame = data->frame; MppTask task = NULL;

if 0

FILE *fpp=fopen("/home/firefly/fw/color3.mjpeg", "rb");
fprintf(stderr,"###start....data->packet_size is %d\n",data->packet_size);
size_t read_size = fread(buf, 1, data->packet_size, fpp);
fprintf(stderr,"####encode1 color image [%0x] [%0x] [%0x] [%0x] [%0x] [%0x]\n",*buf,*(buf+1),*(buf+2),*(buf+3),*(buf+4),*(buf+5));

 if (read_size != data->packet_size || feof(fpp)) {
     mpp_log("found last packet\n");

     // setup eos flag
     data->eos = pkt_eos = 1;
 }

else

size_t read_size = 1843789;
data->eos = pkt_eos = 1;
pkt_eos = 1;
/*if (read_size != data->packet_size ) {
     mpp_log("found last packet\n");

     // setup eos flag
     data->eos = pkt_eos = 1;
 }*/

endif

if 1

 unsigned long long ts = ts_ms();
 memcpy(buf,imageData,imageLen);
 fprintf(stderr, "####first memcpy time is %d\n", (int)(ts_ms()-ts));
 #if 0
 fprintf(stderr,"####encode2 color image [%0x] [%0x] [%0x] [%0x] [%0x] [%0x]\n",*buf,*(buf+1),*(buf+2),*(buf+3),*(buf+4),*(buf+5));
 #endif
 // reset pos
 ts = ts_ms();
 mpp_packet_set_pos(packet, buf);
 mpp_packet_set_length(packet, imageLen);
 fprintf(stderr, "####mpi_dec time1 is %d\n", (int)(ts_ms()-ts));
 // setup eos flag
 if (pkt_eos)
     mpp_packet_set_eos(packet);

endif

        ts = ts_ms();
     ret = mpi->poll(ctx, MPP_PORT_INPUT, MPP_POLL_BLOCK);
     if (ret) {
         mpp_err("mpp input poll failed\n");
         return ret;
     }
     fprintf(stderr, "####mpi_dec time2 is %d\n", (int)(ts_ms()-ts));

 ts = ts_ms();
 ret = mpi->dequeue(ctx, MPP_PORT_INPUT, &task);  /* input queue */
 if (ret) {
     mpp_err("mpp task input dequeue failed\n");
     return ret;
 }
 fprintf(stderr, "####mpi_dec time3 is %d\n", (int)(ts_ms()-ts));

 ts = ts_ms();
 mpp_assert(task);
 fprintf(stderr, "####mpi_dec time4 is %d\n", (int)(ts_ms()-ts));

 ts = ts_ms();
 mpp_task_meta_set_packet(task, KEY_INPUT_PACKET, packet);
 mpp_task_meta_set_frame (task, KEY_OUTPUT_FRAME,  frame);
 fprintf(stderr, "####mpi_dec time5 is %d\n", (int)(ts_ms()-ts));

     ts = ts_ms();
 ret = mpi->enqueue(ctx, MPP_PORT_INPUT, task);  /* input queue */
 if (ret) {
     mpp_err("mpp task input enqueue failed\n");
     return ret;
 }
 fprintf(stderr, "####mpi_dec time6 is %d\n", (int)(ts_ms()-ts));

    /* poll and wait here */
     ts = ts_ms();
     ret = mpi->poll(ctx, MPP_PORT_OUTPUT, MPP_POLL_BLOCK);
     if (ret) {
         mpp_err("mpp output poll failed\n");
         return ret;
     }
     fprintf(stderr, "####mpi_dec time7 is %d\n", (int)(ts_ms()-ts));

     ts = ts_ms();
     ret = mpi->dequeue(ctx, MPP_PORT_OUTPUT, &task); /* output queue */
     if (ret) {
         mpp_err("mpp task output dequeue failed\n");
         return ret;
     }
     fprintf(stderr, "####mpi_dec time8 is %d\n", (int)(ts_ms()-ts));

 mpp_assert(task);

 if (task) {
     MppFrame frame_out = NULL;
     mpp_task_meta_get_frame(task, KEY_OUTPUT_FRAME, &frame_out);
     //mpp_assert(packet_out == packet);

     if (frame) {
         /* write frame to file here */
         MppBuffer buf_out = mpp_frame_get_buffer(frame_out);

         if (buf_out) {
             void *ptr = mpp_buffer_get_ptr(buf_out);
             size_t len  = mpp_buffer_get_size(buf_out);
             ts = ts_ms();
             memcpy(returnData,ptr,len);
             fprintf(stderr, "####sencond memcpy time is %d\n", (int)(ts_ms()-ts));
             //returnData = ptr;
             *returnLen = len;

if 0

             if (data->fp_output)
                 fwrite(ptr, 1, len, data->fp_output);

endif

            #if 0
            mpp_log("####decoded frame %d size %d\n", data->frame_count, len);
            #endif
         }

         if (mpp_frame_get_eos(frame_out))
             mpp_log("found eos frame\n");

     }

     /* output queue */

        #if 1
     ret = mpi->enqueue(ctx, MPP_PORT_OUTPUT, task);
     if (ret)
         mpp_err("mpp task output enqueue failed\n");
     #endif

     //firstrun = 0;
 }

 return ret;

}

MinZhaoLin commented 5 years ago

@HermanChen 你好，请问如何做流水处理？简单地加一个while循环就可以了？我需要进行解码再编码，做了流水后，发现解码和编码的耗时都增加了，难道不能同时进行解码和编码吗？还是需要注意哪里代码的使用？

lwdddddddd commented 5 years ago

你好，我在rk3288平台4.4.55内核上面运行最新的mpp解码，也会遇到上面那个问题，卡在以下部分： HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_gen_regs enter HAL_JPEGD_COMMON: jpegd_setup_output_fmt enter HAL_JPEGD_COMMON: jpegd_setup_output_fmt exit HAL_JPEGD_VDPU1: jpegd_gen_regs enter HAL_JPEGD_VDPU1: jpegd_write_code_word_number enter HAL_JPEGD_VDPU1: jpegd_write_code_word_number exit HAL_JPEGD_COMMON: jpegd_write_qp_ac_dc_table enter HAL_JPEGD_COMMON: jpegd_write_qp_ac_dc_table exit HAL_JPEGD_VDPU1: jpegd_set_chroma_table_id enter HAL_JPEGD_VDPU1: jpegd_set_chroma_table_id exit HAL_JPEGD_VDPU1: jpegd_set_stream_offset enter HAL_JPEGD_VDPU1: jpegd_set_stream_offset exit HAL_JPEGD_VDPU1: jpegd_setup_pp enter HAL_JPEGD_VDPU1: output_frame_fd:0, reg14:7f800000 HAL_JPEGD_VDPU1: jpegd_setup_pp exit HAL_JPEGD_VDPU1: jpegd_gen_regs exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_gen_regs exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_start enter 而我之前在目前最新版本内核4.4.154上面运行mpp解码则会报如下错误： mpp_device: mpp_device_send_reg ioctl VPU_IOC_SET_REG failed ret -1 errno 14 Bad address HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_start mpp_device_send_reg Failed!!! HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_wait enter mpp_device: mpp_device_wait_reg ioctl VPU_IOC_GET_REG failed ret -1 errno 110 Connection timed out HAL_JPEGD_VDPU1: frame_00 output YUV(1920*1088) saving to /tmp/output00.yuv HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_wait exit error: fl_hw_decode failed. mpp_task_impl: mpp_task_queue_deinit idx 0 task 0x7fd5d440 status 3 meta size 0 mpp_task_impl: Assertion p->tasks[i].status == MPP_INPUT_PORT || p->tasks[i].status == MPP_INPUT_HOLD failed at mpp_task_queue_deinit:445 jpegd_parser: jpegd_deinit enter jpegd_parser: jpegd_deinit exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_deinit enter HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_deinit exit 请问你们有测试过的可以用当前版本mpp正常解码的内核版本有哪些？

HermanChen commented 5 years ago

@HermanChen 你好，请问如何做流水处理？简单地加一个while循环就可以了？我需要进行解码再编码，做了流水后，发现解码和编码的耗时都增加了，难道不能同时进行解码和编码吗？还是需要注意哪里代码的使用？

流水是说做多线程处理。不过，jpeg 解码和编码使用的的同一个硬件，多线程处理没什么收益。

HermanChen commented 5 years ago

你好，我在rk3288平台4.4.55内核上面运行最新的mpp解码，也会遇到上面那个问题，卡在以下部分： HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_gen_regs enter HAL_JPEGD_COMMON: jpegd_setup_output_fmt enter HAL_JPEGD_COMMON: jpegd_setup_output_fmt exit HAL_JPEGD_VDPU1: jpegd_gen_regs enter HAL_JPEGD_VDPU1: jpegd_write_code_word_number enter HAL_JPEGD_VDPU1: jpegd_write_code_word_number exit HAL_JPEGD_COMMON: jpegd_write_qp_ac_dc_table enter HAL_JPEGD_COMMON: jpegd_write_qp_ac_dc_table exit HAL_JPEGD_VDPU1: jpegd_set_chroma_table_id enter HAL_JPEGD_VDPU1: jpegd_set_chroma_table_id exit HAL_JPEGD_VDPU1: jpegd_set_stream_offset enter HAL_JPEGD_VDPU1: jpegd_set_stream_offset exit HAL_JPEGD_VDPU1: jpegd_setup_pp enter HAL_JPEGD_VDPU1: output_frame_fd:0, reg14:7f800000 HAL_JPEGD_VDPU1: jpegd_setup_pp exit HAL_JPEGD_VDPU1: jpegd_gen_regs exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_gen_regs exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_start enter 而我之前在目前最新版本内核4.4.154上面运行mpp解码则会报如下错误： mpp_device: mpp_device_send_reg ioctl VPU_IOC_SET_REG failed ret -1 errno 14 Bad address HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_start mpp_device_send_reg Failed!!! HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_wait enter mpp_device: mpp_device_wait_reg ioctl VPU_IOC_GET_REG failed ret -1 errno 110 Connection timed out HAL_JPEGD_VDPU1: frame_00 output YUV(1920*1088) saving to /tmp/output00.yuv HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_wait exit error: fl_hw_decode failed. mpp_task_impl: mpp_task_queue_deinit idx 0 task 0x7fd5d440 status 3 meta size 0 mpp_task_impl: Assertion p->tasks[i].status == MPP_INPUT_PORT || p->tasks[i].status == MPP_INPUT_HOLD failed at mpp_task_queue_deinit:445 jpegd_parser: jpegd_deinit enter jpegd_parser: jpegd_deinit exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_deinit enter HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_deinit exit 请问你们有测试过的可以用当前版本mpp正常解码的内核版本有哪些？

遇到这种情况，先运行下 mpp_platform_test 和 mpp_runtime_test 看一下。

lwdddddddd commented 5 years ago

你好，我在rk3288平台4.4.55内核上面运行最新的mpp解码，也会遇到上面那个问题，卡在以下部分： HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_gen_regs enter HAL_JPEGD_COMMON: jpegd_setup_output_fmt enter HAL_JPEGD_COMMON: jpegd_setup_output_fmt exit HAL_JPEGD_VDPU1: jpegd_gen_regs enter HAL_JPEGD_VDPU1: jpegd_write_code_word_number enter HAL_JPEGD_VDPU1: jpegd_write_code_word_number exit HAL_JPEGD_COMMON: jpegd_write_qp_ac_dc_table enter HAL_JPEGD_COMMON: jpegd_write_qp_ac_dc_table exit HAL_JPEGD_VDPU1: jpegd_set_chroma_table_id enter HAL_JPEGD_VDPU1: jpegd_set_chroma_table_id exit HAL_JPEGD_VDPU1: jpegd_set_stream_offset enter HAL_JPEGD_VDPU1: jpegd_set_stream_offset exit HAL_JPEGD_VDPU1: jpegd_setup_pp enter HAL_JPEGD_VDPU1: output_frame_fd:0, reg14:7f800000 HAL_JPEGD_VDPU1: jpegd_setup_pp exit HAL_JPEGD_VDPU1: jpegd_gen_regs exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_gen_regs exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_start enter 而我之前在目前最新版本内核4.4.154上面运行mpp解码则会报如下错误： mpp_device: mpp_device_send_reg ioctl VPU_IOC_SET_REG failed ret -1 errno 14 Bad address HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_start mpp_device_send_reg Failed!!! HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_wait enter mpp_device: mpp_device_wait_reg ioctl VPU_IOC_GET_REG failed ret -1 errno 110 Connection timed out HAL_JPEGD_VDPU1: frame_00 output YUV(1920*1088) saving to /tmp/output00.yuv HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_wait exit error: fl_hw_decode failed. mpp_task_impl: mpp_task_queue_deinit idx 0 task 0x7fd5d440 status 3 meta size 0 mpp_task_impl: Assertion p->tasks[i].status == MPP_INPUT_PORT || p->tasks[i].status == MPP_INPUT_HOLD failed at mpp_task_queue_deinit:445 jpegd_parser: jpegd_deinit enter jpegd_parser: jpegd_deinit exit HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_deinit enter HAL_JPEGD_VDPU1: hal_jpegd_vdpu1_deinit exit 请问你们有测试过的可以用当前版本mpp正常解码的内核版本有哪些？

遇到这种情况，先运行下 mpp_platform_test 和 mpp_runtime_test 看一下。

我在4.4.55和4.4.154上面都运行了，结果是正常的，下面是打印输出： root@linaro-alip:/home/vvwork/mpp/mpp-release/osal/test# ./mpp_platform_test mpp_plat_test: chip name: rockchip,rk3288w mpp_plat_test: mpp_plat_test: chip vcodec type 00000101 mpp_plat_test: found vpu1 codec mpp_plat_test: found RK hevc decoder mpp_plat_test: mpp_plat_test: start probing decoder device name: mpp_plat_test: H.264 decoder: /dev/vpu-service mpp_plat_test: H.265 decoder: /dev/hevc-service mpp_plat_test: MJPEG decoder: /dev/vpu-service mpp_plat_test: VP9 decoder: /dev/vpu-service mpp_plat_test: avs decoder: /dev/vpu-service mpp_plat_test: mpp_plat_test: start probing encoder device name: mpp_plat_test: H.264 encoder: /dev/vpu-service mpp_plat_test: H.265 encoder: /dev/vpu-service mpp_plat_test: MJPEG encoder: /dev/vpu-service mpp_plat_test: mpp platform test done root@linaro-alip:/home/vvwork/mpp/mpp-release/osal/test# ./mpp_runtime_test mpp_rt: NOT found ion allocator mpp_rt: found drm allocator mpp_rt_test: mpp found ion buffer is invalid mpp_rt_test: mpp found drm buffer is valid

lwdddddddd commented 5 years ago

@HermanChen 你好，我安装了一下firefly-rk3288最新的ubuntu16.04镜像，内核版本也是4.4.154。我发现系统自带的mpp是可以正常解码的，不会报如上VPU错误，另外循环解码也没有问题。系统自带的mpp版本是：mpi: mpp version: 10c9a2d6 author: Herman Chen [jpegd]: Add vpu jpegd stream patch。而会报错的版本是：mpi: mpp version: Without VCS info，这个是我下载的最新的mpp编译出来的。

HermanChen commented 5 years ago

Without VCS info 这个说明没有版本管理信息，可能是非标版本？

HermanChen commented 5 years ago

下最新 mpp 的话，应该是会带有 git 信息的

lwdddddddd commented 5 years ago

下最新 mpp 的话，应该是会带有 git 信息的

我重新下载编译了一下，还是会报错，版本信息是：mpi: mpp version: 3d353987 author: Johnson Ding [jpege]: Fix jpeg encoder stride problem

Archer112 commented 4 years ago

下最新mpp的话，应该是会带有git信息的

你好 我使用这个demo 做循环解码jpg图片 运行不到一分钟出现了这个错误 请问是什么原因 pi_dec_test start input file size 960000 mpi_dec_test decoder test start w 1280 h 720 type 8 mpi: mpp version: 0b701fe author: Lawrence-Tang DEBIAN: prepare for release_20171218-3 mpp_device: mpp_device_init failed to open device /dev/vpu_service, errno 24, error msg: Too many open files HAL_JPEG_VDPU2: hal_jpegd_vdpu2_init get vpu_socket(-1) <= 0, failed. mpp_hal: mpp_hal_init hal jpegd init failed ret -2 mpp_hal: mpp_hal_init could not found coding type 8 mpp_dec: mpp_dec_init could not init hal mpp: error found on mpp initialization mpp_dec: mpp_dec_control found NULL input dec (nil) mpp: command 31000a param 0x7fce15fc04 ret -3 mpp input poll failed found last packet mpp input poll failed mpi->reset failed