Hi3559AV100 SVP NNIE 功能仿真及CUDA加速

最新推荐文章于 2024-04-12 21:48:38 发布
最新推荐文章于 2024-04-12 21:48:38 发布
阅读量4.7k 收藏 5
点赞数 2
分类专栏： 海思深度学习加速 文章标签： 海思深度学习加速
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/u012215960/article/details/100638895
版权

海思的Hi3559AV100提供在Ruyi Studio中进行仿真的能力。

仿真的CUDA加速

功能仿真中可以使用CUDA加速配置。配置方式如下。

环境要求

Windows版本 7/10，CUDA需要 >= 8.0，显存需要 >= 6G
从NVIDIA主页下载CUDA
https://developer.nvidia.com/cuda-downloads?target_os=Windows&target_arch=x86_64

架构支持

SM_30, SM_35
SM_50, SM_52, SM_53
SM_60, SM_61, SM_62

启动仿真

仿真默认设置

仿真设置在一个绝对路径
Windows：C:\hisilicon\nnie_sim.ini
Linux：/usr/hisilicon/nnie_sim.ini
默认配置是

#===== CONFIG =====
[LAYER_LINEAR_PRINT_EN]
0
[CUDA_CALC_EN]
0

跑默认的sample.exe产生console如下

Sun Sep  8 22:18:21 2019 E config openfile failure(path: C:/hisilicon/nnie_sim.ini)
Sun Sep  8 22:18:21 2019 W C:/hisilicon/nnie_sim.ini load failure, using default setting
Sun Sep  8 22:18:21 2019 I ============= nnie_sim.ini setting ==============
Sun Sep  8 22:18:21 2019 I [LOG_LEVEL]                      DBG
Sun Sep  8 22:18:21 2019 I =================================================
SvpSampleCnnClfLenet start ...
SVP Version: HiSVP_PC_V1.1.1.0_B030
build sha1: 86265e2c
      date: Jun 20 2018
      time: 15:41:12
Read version info from WK header
      sdk_arch_ver: 	1.1
      nnie_arch_ver: 	1.1
      nnie_mapper_ver: 	1.1.1.0
      test_ver: 	B030
      patch_ver: 	P000
Sun Sep  8 22:18:30 2019 I >>>>>>>>> createNet >>>>>>>>> virAddr = 0x1743fd0
Sun Sep  8 22:18:30 2019 D MAIN HI_MPI_SVP_NNIE_LoadModel():152 net ptr addr: 000000000031F740
Sun Sep  8 22:18:30 2019 D NET Net::setDumpPath():385 [DUMP_INDIVIDUAL_EN] = 0
Sun Sep  8 22:18:30 2019 D NET Net::setDumpPath():401 set net DumpPath: ./
Sun Sep  8 22:18:30 2019 D NET Net::setUp():255 model net base: 0000000001743FD0, segment num: 1
Sun Sep  8 22:18:30 2019 D NET Net::setUp():263 model seg(0) base: 0000000001744010
Sun Sep  8 22:18:30 2019 I create net seg(0), type: 205
Sun Sep  8 22:18:30 2019 D SEG Segment::DumpParamHdr():470 netType: 0 srcNodeNum: 1, dstNodeNum: 1
Sun Sep  8 22:18:30 2019 D SEG Segment::DumpParamHdr():471 tmpBufSize: 4734976, netSegSize: 96, netSegLayerNum: 9
Sun Sep  8 22:18:30 2019 D SEG Segment::DumpParamHdr():472 maxT: 0
Sun Sep  8 22:18:30 2019 D SEG Segment::DumpSrcNodeInfo():483 srcNode width: 28, height: 28, channel: 1, fmt: 1, isvec: 0, nodeid: 0
Sun Sep  8 22:18:30 2019 D SEG Segment::DumpDstNodeInfo():495 dstNode width: 1, height: 1, channel: 10, fmt: 0,  isvec: 1, nodeid: 8
Sun Sep  8 22:18:30 2019 D SEG Segment::setUp():435 model seg base: 0000000001744090, layers num: 8
Sun Sep  8 22:18:30 2019 D SEG Segment::setUp():442 model layer base: 0x1744090
Sun Sep  8 22:18:30 2019 I create layer(0): preprocess
Sun Sep  8 22:18:30 2019 D DATA DataLayer::setUp():139 param size: 64
Sun Sep  8 22:18:30 2019 D SEG Segment::setUp():442 model layer base: 0x17440d0
Sun Sep  8 22:18:30 2019 I create layer(1): convolution
Sun Sep  8 22:18:30 2019 D CONV ConvolutionLayer::setUp():126 param size: 128
Sun Sep  8 22:18:30 2019 D SEG Segment::setUp():442 model layer base: 0x1744150
Sun Sep  8 22:18:30 2019 I create layer(2): poolingave
Sun Sep  8 22:18:30 2019 D MAXP PoolingLayer::setUp():27 param size: 96
Sun Sep  8 22:18:30 2019 D SEG Segment::setUp():442 model layer base: 0x17441b0
Sun Sep  8 22:18:30 2019 I create layer(3): convolution
Sun Sep  8 22:18:30 2019 D CONV ConvolutionLayer::setUp():126 param size: 128

观测GPU利用率

仿真CUDA设置

仿真设置在一个绝对路径
Windows：C:\hisilicon\nnie_sim.ini
Linux：/usr/hisilicon/nnie_sim.ini
默认配置是

#===== CONFIG =====
[LAYER_LINEAR_PRINT_EN]
0
[CUDA_CALC_EN]
1

跑默认的sample.exe产生console如下

Sun Sep  8 22:46:50 2019 I ============= nnie_sim.ini setting ==============
Sun Sep  8 22:46:50 2019 I [CUDA_CALC_EN]                   1
Sun Sep  8 22:46:50 2019 I [LOG_LEVEL]                      DBG
Sun Sep  8 22:46:50 2019 I =================================================
SvpSampleCnnClfLenet start ...
SVP Version: HiSVP_PC_V1.1.1.0_B030
build sha1: 86265e2c
      date: Jun 20 2018
      time: 15:41:12
Read version info from WK header
      sdk_arch_ver: 	1.1
      nnie_arch_ver: 	1.1
      nnie_mapper_ver: 	1.1.1.0
      test_ver: 	B030
      patch_ver: 	P000
Sun Sep  8 22:46:55 2019 I >>>>>>>>> createNet >>>>>>>>> virAddr = 0x2ed3fd0
Sun Sep  8 22:46:55 2019 D CUDA cudaHostManager::loadLibray():88 LoadLibrary nnieCUDA1.1d.dll finished
Sun Sep  8 22:46:55 2019 D CUDA cudaHostManager::versionCheck():136 lib version check: nnieCUDA = HiSVP_PC_V1.1.1.0_B030, nnie Simulator = HiSVP_PC_V1.1.1.0_B030
                          CUDA Device(0) Properties
Identify: Quadro M2000
Clock Rate: 1162500 kHz
Max Grid Size: 2147483647 * 65535 * 65535
Max Threads Dim: 1024 * 1024 * 64
Max Threads per Block: 1024
Number of Multiprocessors: 6
32bit Registers Available per Block: 65536
Warp Size: 32 threads
Sun Sep  8 22:46:56 2019 D CUDA NetManager::createNet():45 nnieCUDA create success
Sun Sep  8 22:46:56 2019 D MAIN HI_MPI_SVP_NNIE_LoadModel():152 net ptr addr: 000000000031F870
Sun Sep  8 22:46:56 2019 D NET Net::setDumpPath():385 [DUMP_INDIVIDUAL_EN] = 0
Sun Sep  8 22:46:56 2019 D NET Net::setDumpPath():401 set net DumpPath: ./
Sun Sep  8 22:46:56 2019 D NET Net::setUp():255 model net base: 0000000002ED3FD0, segment num: 1
Sun Sep  8 22:46:56 2019 D NET Net::setUp():263 model seg(0) base: 0000000002ED4010
Sun Sep  8 22:46:56 2019 I create net seg(0), type: 205
Sun Sep  8 22:46:56 2019 D SEG Segment::DumpParamHdr():470 netType: 0 srcNodeNum: 1, dstNodeNum: 1
Sun Sep  8 22:46:56 2019 D SEG Segment::DumpParamHdr():471 tmpBufSize: 4734976, netSegSize: 96, netSegLayerNum: 9
Sun Sep  8 22:46:56 2019 D SEG Segment::DumpParamHdr():472 maxT: 0
Sun Sep  8 22:46:56 2019 D SEG Segment::DumpSrcNodeInfo():483 srcNode width: 28, height: 28, channel: 1, fmt: 1, isvec: 0, nodeid: 0
Sun Sep  8 22:46:56 2019 D SEG Segment::DumpDstNodeInfo():495 dstNode width: 1, height: 1, channel: 10, fmt: 0,  isvec: 1, nodeid: 8
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():435 model seg base: 0000000002ED4090, layers num: 8
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():442 model layer base: 0x2ed4090
Sun Sep  8 22:46:56 2019 I create layer(0): preprocess
Sun Sep  8 22:46:56 2019 D DATA DataLayer::setUp():139 param size: 64
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():442 model layer base: 0x2ed40d0
Sun Sep  8 22:46:56 2019 I create layer(1): convolution
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setUp():126 param size: 128
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():442 model layer base: 0x2ed4150
Sun Sep  8 22:46:56 2019 I create layer(2): poolingave
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setUp():27 param size: 96
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():442 model layer base: 0x2ed41b0
Sun Sep  8 22:46:56 2019 I create layer(3): convolution
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setUp():126 param size: 128
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():442 model layer base: 0x2ed4230
Sun Sep  8 22:46:56 2019 I create layer(4): poolingave
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setUp():27 param size: 96
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():442 model layer base: 0x2ed4290
Sun Sep  8 22:46:56 2019 I create layer(5): innerproduct
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setUp():24 param size: 112
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():442 model layer base: 0x2ed4300
Sun Sep  8 22:46:56 2019 I create layer(6): innerproduct
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setUp():24 param size: 112
Sun Sep  8 22:46:56 2019 D SEG Segment::setUp():442 model layer base: 0x2ed4370
Sun Sep  8 22:46:56 2019 I create layer(7): softmax
Sun Sep  8 22:46:56 2019 D SOFTMAX SoftmaxLayer::setUp():55 param size: 64
Sun Sep  8 22:46:56 2019 D NET Net::setUp():274 seg(0) all param size: 0x3a0
Sun Sep  8 22:46:56 2019 D MAIN HI_MPI_SVP_NNIE_LoadModel():162 paramBuf addr: 0000000002ED43B0, size: 433824
Sun Sep  8 22:46:56 2019 D NET Net::paramWKInit():110 +
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000002ED43B0, size: 0x69ea0
Sun Sep  8 22:46:56 2019 D NET Net::paramWKInit():110 -
Sun Sep  8 22:46:56 2019 D MAIN HI_MPI_SVP_NNIE_Forward():902 net ptr addr: 000000000031F870
Sun Sep  8 22:46:56 2019 D NET Net::tmpWKInit():109 +
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004660040, size: 0x484000
Sun Sep  8 22:46:56 2019 D NET Net::tmpWKInit():109 -
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000192A430, size: 0x40
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000192A470, size: 0x20
Sun Sep  8 22:46:56 2019 D NET Net::inputWKsInit():61 +
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000192C5C0, size: 0x380
Sun Sep  8 22:46:56 2019 D NET Net::inputWKsInit():114 -
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000192C970, size: 0x30
Sun Sep  8 22:46:56 2019 D SEG CNNSegment::forward():10 compute begin
Sun Sep  8 22:46:56 2019 I start [seg(0) layer(0)]: preprocess
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000192C5C0, size: 0x380
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004660040, size: 0x310
Sun Sep  8 22:46:56 2019 D DATA DataLayer::setAddr():167 [0] inUsrOffset: 0, input base: 0x192c5c0
Sun Sep  8 22:46:56 2019 D DATA DataLayer::setAddr():168 [0] outOffset: 0, output base: 0x4660040
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer0_output0_func
Sun Sep  8 22:46:56 2019 I end   [seg(0) layer(0)]: preprocess
Sun Sep  8 22:46:56 2019 I start [seg(0) layer(1)]: convolution
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004660040, size: 0x310
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004698040, size: 0x2d00
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002ED43B0, size: 0
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002ED43B0, size: 0
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000002ED43B0, size: 0x1f4
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002ED45A4, size: 0x50
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002ED45F4, size: 0
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():279 [1] inOffset: 0, input base: 0
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():280 [1] outOffset: 0, output base: 0x192c970
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():284 [1] input range(784): [0x4660040, 0x4660040]
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():294 [1] output range(11520): [0x4698040, 0x469ad40]
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():302 [1] bnDeltaA param range(0): [0x2ed43b0, 0x2ed43b0], offset: 0
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():304 [1] bnDeltB param range(0): [0x2ed43b0, 0x2ed43b0], offset: 0
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():306 [1] weight param range(500): [0x2ed43b0, 0x2ed45a4], offset: 0
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():308 [1] bias param range(20): [0x2ed45a4, 0x2ed45f4], offset: 0x1f4
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():311 [1] preluNegtiveSlope param range(0): [0x2ed45f4, 0x2ed45f4], offset: 0x244
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::compute():393 GPU conv
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer1_output0_func
Sun Sep  8 22:46:56 2019 I end   [seg(0) layer(1)]: convolution
Sun Sep  8 22:46:56 2019 I start [seg(0) layer(2)]: poolingave
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004698040, size: 0x2d00
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004968040, size: 0xb40
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setAddr():92 [2] outOffset: 3178496, output base: 0x192c970
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setAddr():93 [2] inOffset: 229376, input base: 0
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setAddr():97 [2] input range(11520): [0x4698040, 0x469ad40]
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setAddr():107 [2] output range(2880): [0x4968040, 0x4968b80]
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer2_output0_func
Sun Sep  8 22:46:56 2019 I end   [seg(0) layer(2)]: poolingave
Sun Sep  8 22:46:56 2019 I start [seg(0) layer(3)]: convolution
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004968040, size: 0xb40
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004660040, size: 0xc80
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002ED4600, size: 0
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002ED4600, size: 0
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000002ED4600, size: 0x61a8
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002EDA7A8, size: 0xc8
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002EDA870, size: 0
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():279 [3] inOffset: 3178496, input base: 0
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():280 [3] outOffset: 0, output base: 0x192c970
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():284 [3] input range(2880): [0x4968040, 0x4968040]
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():294 [3] output range(3200): [0x4660040, 0x4660cc0]
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():302 [3] bnDeltaA param range(0): [0x2ed4600, 0x2ed4600], offset: 0x250
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():304 [3] bnDeltB param range(0): [0x2ed4600, 0x2ed4600], offset: 0x250
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():306 [3] weight param range(25000): [0x2ed4600, 0x2eda7a8], offset: 0x250
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():308 [3] bias param range(50): [0x2eda7a8, 0x2eda870], offset: 0x63f8
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::setAddr():311 [3] preluNegtiveSlope param range(0): [0x2eda870, 0x2eda870], offset: 0x64c0
Sun Sep  8 22:46:56 2019 D CONV ConvolutionLayer::compute():393 GPU conv
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer3_output0_func
Sun Sep  8 22:46:56 2019 I end   [seg(0) layer(3)]: convolution
Sun Sep  8 22:46:56 2019 I start [seg(0) layer(4)]: poolingave
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004660040, size: 0xc80
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004A1C040, size: 0x320
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setAddr():92 [4] outOffset: 3915776, output base: 0x192c970
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setAddr():93 [4] inOffset: 0, input base: 0
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setAddr():97 [4] input range(3200): [0x4660040, 0x4660cc0]
Sun Sep  8 22:46:56 2019 D MAXP PoolingLayer::setAddr():107 [4] output range(800): [0x4a1c040, 0x4a1c360]
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer4_output0_func
Sun Sep  8 22:46:56 2019 I end   [seg(0) layer(4)]: poolingave
Sun Sep  8 22:46:56 2019 I start [seg(0) layer(5)]: innerproduct
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004A1C040, size: 0x320
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004660040, size: 0x1f4
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000002EDA870, size: 0x61a80
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002F3C2F0, size: 0x7d0
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002F3CAC0, size: 0
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():106 [5] inOffset: 0x3bc000, input base: 0
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():107 [5] outOffset: 0, output base: 0x192c970
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():109 [5] weight param range(400000): [0x2eda870, 0x2f3c2f0], offset: 0x64c0
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():111 [5] bias param range(2000): [0x2f3c2f0, 0x2f3cac0], offset: 0x67f40
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():113 [5] BNDeltaA param range(0): [0, 0], offset: 0x68710
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():115 [5] BNDeltaA param range(?): [0, ?], offset: 0x68710
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():119 [5] input range(800): [0x4a1c040, 0x4a1c360]
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():129 [5] output range(500): [0x4660040, 0x4660234]
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer5_output0_func
Sun Sep  8 22:46:56 2019 I end   [seg(0) layer(5)]: innerproduct
Sun Sep  8 22:46:56 2019 I start [seg(0) layer(6)]: innerproduct
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000004660040, size: 0x1f4
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000477A040, size: 0x28
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000002F3CAC0, size: 0x1388
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002F3DE48, size: 0x28
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<int>::init():175 base: 0000000002F3DE70, size: 0
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():106 [6] inOffset: 0, input base: 0
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():107 [6] outOffset: 0x11a000, output base: 0x192c970
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():109 [6] weight param range(5000): [0x2f3cac0, 0x2f3de48], offset: 0x68710
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():111 [6] bias param range(40): [0x2f3de48, 0x2f3de70], offset: 0x69a98
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():113 [6] BNDeltaA param range(0): [0, 0], offset: 0x69ac0
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():115 [6] BNDeltaA param range(?): [0, ?], offset: 0x69ac0
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():119 [6] input range(500): [0x4660040, 0x4660234]
Sun Sep  8 22:46:56 2019 D IP InnerProductLayer::setAddr():129 [6] output range(10): [0x477a040, 0x477a068]
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer6_output0_func
Sun Sep  8 22:46:56 2019 I end   [seg(0) layer(6)]: innerproduct
Sun Sep  8 22:46:56 2019 I start [seg(0) layer(7)]: softmax
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000477A040, size: 0x28
Sun Sep  8 22:46:56 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000192C970, size: 0x30
Sun Sep  8 22:46:56 2019 D SOFTMAX SoftmaxLayer::setAddr():141 [7] inOffset: 1155072, input base: 0
Sun Sep  8 22:46:56 2019 D SOFTMAX SoftmaxLayer::setAddr():142 [7] outOffset: 0, output base: 0x192c970
Sun Sep  8 22:46:56 2019 D SOFTMAX SoftmaxLayer::setAddr():146 [7] input range(10): [0x477a040, 0x477a068]
Sun Sep  8 22:46:56 2019 D SOFTMAX SoftmaxLayer::setAddr():161 [7] output range(10): [0x192c970, 0x192c9a0]
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:56 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer7_output0_func
Sun Sep  8 22:46:56 2019 I end   [seg(0) layer(7)]: softmax
fseek return value is 0
fseek return value is 0
Leaf0, Pic0 --> expected label: 0
Top0: index --    0, confidence -- 1.0000000
Top1: index --    1, confidence -- 0.0000000
Top2: index --    2, confidence -- 0.0000000
Top3: index --    3, confidence -- 0.0000000
Top4: index --    4, confidence -- 0.0000000
cudaSetDevice(0)
Sun Sep  8 22:46:56 2019 D CUDA cudaHostManager::freeLibray():111 FreeLibrary nnieCUDA1.1d.dll finished
Sun Sep  8 22:46:56 2019 I <<<<<<<<< deleteNet <<<<<<<<< virAddr = 0x2ed3fd0
SvpSampleCnnClfLenet end ...
SvpSampleCnnClfAlexnet start ...
SVP Version: HiSVP_PC_V1.1.1.0_B030
build sha1: 86265e2c
      date: Jun 20 2018
      time: 15:41:12
Read version info from WK header
      sdk_arch_ver: 	1.1
      nnie_arch_ver: 	1.1
      nnie_mapper_ver: 	1.1.1.0
      test_ver: 	B030
      patch_ver: 	P000
Sun Sep  8 22:46:56 2019 I >>>>>>>>> createNet >>>>>>>>> virAddr = 0x75f0040
Sun Sep  8 22:46:56 2019 D CUDA cudaHostManager::loadLibray():88 LoadLibrary nnieCUDA1.1d.dll finished
                          CUDA Device(0) Properties
Identify: Quadro M2000
Clock Rate: 1162500 kHz
Max Grid Size: 2147483647 * 65535 * 65535
Max Threads Dim: 1024 * 1024 * 64
Max Threads per Block: 1024
Number of Multiprocessors: 6
32bit Registers Available per Block: 65536
Warp Size: 32 threads
Sun Sep  8 22:46:56 2019 D CUDA NetManager::createNet():45 nnieCUDA create success
Sun Sep  8 22:46:56 2019 D MAIN HI_MPI_SVP_NNIE_LoadModel():152 net ptr addr: 0000000007222010
Sun Sep  8 22:46:56 2019 D NET Net::setDumpPath():385 [DUMP_INDIVIDUAL_EN] = 0
Sun Sep  8 22:46:56 2019 D NET Net::setDumpPath():401 set net DumpPath: ./
Sun Sep  8 22:46:57 2019 D NET Net::setUp():255 model net base: 00000000075F0040, segment num: 1
Sun Sep  8 22:46:57 2019 D NET Net::setUp():263 model seg(0) base: 00000000075F0080
Sun Sep  8 22:46:57 2019 I create net seg(0), type: 205
Sun Sep  8 22:46:57 2019 D SEG Segment::DumpParamHdr():470 netType: 0 srcNodeNum: 1, dstNodeNum: 1
Sun Sep  8 22:46:57 2019 D SEG Segment::DumpParamHdr():471 tmpBufSize: 221147136, netSegSize: 96, netSegLayerNum: 16
Sun Sep  8 22:46:57 2019 D SEG Segment::DumpParamHdr():472 maxT: 0
Sun Sep  8 22:46:57 2019 D SEG Segment::DumpSrcNodeInfo():483 srcNode width: 227, height: 227, channel: 3, fmt: 1, isvec: 0, nodeid: 0
Sun Sep  8 22:46:57 2019 D SEG Segment::DumpDstNodeInfo():495 dstNode width: 1, height: 1, channel: 1000, fmt: 0,  isvec: 1, nodeid: 15
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():435 model seg base: 00000000075F0100, layers num: 15
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0100
Sun Sep  8 22:46:57 2019 I create layer(0): preprocess
Sun Sep  8 22:46:57 2019 D DATA DataLayer::setUp():139 param size: 64
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0140
Sun Sep  8 22:46:57 2019 I create layer(1): convolution
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setUp():126 param size: 128
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f01c0
Sun Sep  8 22:46:57 2019 I create layer(2): lrn
Sun Sep  8 22:46:57 2019 D LRN LrnLayer::setUp():77 param size: 80
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0210
Sun Sep  8 22:46:57 2019 I create layer(3): poolingave
Sun Sep  8 22:46:57 2019 D MAXP PoolingLayer::setUp():27 param size: 96
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0270
Sun Sep  8 22:46:57 2019 I create layer(4): convolution
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setUp():126 param size: 128
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f02f0
Sun Sep  8 22:46:57 2019 I create layer(5): lrn
Sun Sep  8 22:46:57 2019 D LRN LrnLayer::setUp():77 param size: 80
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0340
Sun Sep  8 22:46:57 2019 I create layer(6): poolingave
Sun Sep  8 22:46:57 2019 D MAXP PoolingLayer::setUp():27 param size: 96
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f03a0
Sun Sep  8 22:46:57 2019 I create layer(7): convolution
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setUp():126 param size: 128
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0420
Sun Sep  8 22:46:57 2019 I create layer(8): convolution
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setUp():126 param size: 128
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f04a0
Sun Sep  8 22:46:57 2019 I create layer(9): convolution
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setUp():126 param size: 128
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0520
Sun Sep  8 22:46:57 2019 I create layer(10): poolingave
Sun Sep  8 22:46:57 2019 D MAXP PoolingLayer::setUp():27 param size: 96
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0580
Sun Sep  8 22:46:57 2019 I create layer(11): innerproduct
Sun Sep  8 22:46:57 2019 D IP InnerProductLayer::setUp():24 param size: 112
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f05f0
Sun Sep  8 22:46:57 2019 I create layer(12): innerproduct
Sun Sep  8 22:46:57 2019 D IP InnerProductLayer::setUp():24 param size: 112
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f0660
Sun Sep  8 22:46:57 2019 I create layer(13): innerproduct
Sun Sep  8 22:46:57 2019 D IP InnerProductLayer::setUp():24 param size: 112
Sun Sep  8 22:46:57 2019 D SEG Segment::setUp():442 model layer base: 0x75f06d0
Sun Sep  8 22:46:57 2019 I create layer(14): softmax
Sun Sep  8 22:46:57 2019 D SOFTMAX SoftmaxLayer::setUp():55 param size: 64
Sun Sep  8 22:46:57 2019 D NET Net::setUp():274 seg(0) all param size: 0x690
Sun Sep  8 22:46:57 2019 D MAIN HI_MPI_SVP_NNIE_LoadModel():162 paramBuf addr: 00000000075F0710, size: 62411792
Sun Sep  8 22:46:57 2019 D NET Net::paramWKInit():110 +
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 00000000075F0710, size: 0x3b85410
Sun Sep  8 22:46:57 2019 D NET Net::paramWKInit():110 -
Sun Sep  8 22:46:57 2019 D MAIN HI_MPI_SVP_NNIE_Forward():902 net ptr addr: 0000000007222010
Sun Sep  8 22:46:57 2019 D NET Net::tmpWKInit():109 +
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000B180040, size: 0xd2e7000
Sun Sep  8 22:46:57 2019 D NET Net::tmpWKInit():109 -
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000192CBA0, size: 0x40
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000192CBE0, size: 0x30
Sun Sep  8 22:46:57 2019 D NET Net::inputWKsInit():61 +
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000002ED3FD0, size: 0x27e70
Sun Sep  8 22:46:57 2019 D NET Net::inputWKsInit():114 -
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000002EFBE70, size: 0xfa0
Sun Sep  8 22:46:57 2019 D SEG CNNSegment::forward():10 compute begin
Sun Sep  8 22:46:57 2019 I start [seg(0) layer(0)]: preprocess
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000002ED3FD0, size: 0x27e70
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000B180040, size: 0x25bdb
Sun Sep  8 22:46:57 2019 D DATA DataLayer::setAddr():167 [0] inUsrOffset: 0, input base: 0x2ed3fd0
Sun Sep  8 22:46:57 2019 D DATA DataLayer::setAddr():168 [0] outOffset: 0, output base: 0xb180040
Sun Sep  8 22:46:57 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:57 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer0_output0_func
Sun Sep  8 22:46:57 2019 I end   [seg(0) layer(0)]: preprocess
Sun Sep  8 22:46:57 2019 I start [seg(0) layer(1)]: convolution
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000B180040, size: 0x25bdb
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000D967040, size: 0x46e60
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<int>::init():175 base: 00000000075F0710, size: 0
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<int>::init():175 base: 00000000075F0710, size: 0
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 00000000075F0710, size: 0x8820
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<int>::init():175 base: 00000000075F8F30, size: 0x180
Sun Sep  8 22:46:57 2019 D MAIN MemoryWK<int>::init():175 base: 00000000075F90B0, size: 0
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():279 [1] inOffset: 0, input base: 0
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():280 [1] outOffset: 0, output base: 0x2efbe70
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():284 [1] input range(154587): [0xb180040, 0xb180040]
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():294 [1] output range(290400): [0xd967040, 0xd9adea0]
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():302 [1] bnDeltaA param range(0): [0x75f0710, 0x75f0710], offset: 0
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():304 [1] bnDeltB param range(0): [0x75f0710, 0x75f0710], offset: 0
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():306 [1] weight param range(34848): [0x75f0710, 0x75f8f30], offset: 0
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():308 [1] bias param range(96): [0x75f8f30, 0x75f90b0], offset: 0x8820
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::setAddr():311 [1] preluNegtiveSlope param range(0): [0x75f90b0, 0x75f90b0], offset: 0x89a0
Sun Sep  8 22:46:57 2019 D CONV ConvolutionLayer::compute():393 GPU conv
Sun Sep  8 22:46:58 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:58 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer1_output0_func
Sun Sep  8 22:46:58 2019 I end   [seg(0) layer(1)]: convolution
Sun Sep  8 22:46:58 2019 I start [seg(0) layer(2)]: lrn
Sun Sep  8 22:46:58 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000D967040, size: 0x46e60
Sun Sep  8 22:46:58 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000012BE7040, size: 0x46e60
Sun Sep  8 22:46:58 2019 D LRN LrnLayer::setAddr():158 [2] inOffset: 41840640, input base: 0
Sun Sep  8 22:46:58 2019 D LRN LrnLayer::setAddr():159 [2] outOffset: 128348160, output base: 0x2efbe70
Sun Sep  8 22:46:58 2019 D LRN LrnLayer::setAddr():163 [2] input range(290400): [0xd967040, 0xd967040]
Sun Sep  8 22:46:58 2019 D LRN LrnLayer::setAddr():173 [2] output range(290400): [0x12be7040, 0x12be7040]
Sun Sep  8 22:46:59 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:59 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer2_output0_func
Sun Sep  8 22:46:59 2019 I end   [seg(0) layer(2)]: lrn
Sun Sep  8 22:46:59 2019 I start [seg(0) layer(3)]: poolingave
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 0000000012BE7040, size: 0x46e60
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000B180040, size: 0x11160
Sun Sep  8 22:46:59 2019 D MAXP PoolingLayer::setAddr():92 [3] outOffset: 0, output base: 0x2efbe70
Sun Sep  8 22:46:59 2019 D MAXP PoolingLayer::setAddr():93 [3] inOffset: 128348160, input base: 0
Sun Sep  8 22:46:59 2019 D MAXP PoolingLayer::setAddr():97 [3] input range(290400): [0x12be7040, 0x12c2dea0]
Sun Sep  8 22:46:59 2019 D MAXP PoolingLayer::setAddr():107 [3] output range(69984): [0xb180040, 0xb1911a0]
Sun Sep  8 22:46:59 2019 D NET tools::dumpLayerBase():447 dumpLayerBase Path: ./
Sun Sep  8 22:46:59 2019 D NET tools::dumpLayerBase():448 dumpLayerBase filename: seg0_layer3_output0_func
Sun Sep  8 22:46:59 2019 I end   [seg(0) layer(3)]: poolingave
Sun Sep  8 22:46:59 2019 I start [seg(0) layer(4)]: convolution
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000B180040, size: 0x11160
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 000000000C338040, size: 0x2d900
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<int>::init():175 base: 00000000075F90B0, size: 0
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<int>::init():175 base: 00000000075F90B0, size: 0
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<unsigned char>::init():175 base: 00000000075F90B0, size: 0x96000
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<int>::init():175 base: 000000000768F0B0, size: 0x400
Sun Sep  8 22:46:59 2019 D MAIN MemoryWK<int>::init():175 base: 000000000768F4B0, size: 0
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():279 [4] inOffset: 0, input base: 0
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():280 [4] outOffset: 0, output base: 0x2efbe70
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():284 [4] input range(69984): [0xb180040, 0xb180040]
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():294 [4] output range(186624): [0xc338040, 0xc365940]
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():302 [4] bnDeltaA param range(0): [0x75f90b0, 0x75f90b0], offset: 0x89a0
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():304 [4] bnDeltB param range(0): [0x75f90b0, 0x75f90b0], offset: 0x89a0
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():306 [4] weight param range(614400): [0x75f90b0, 0x768f0b0], offset: 0x89a0
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():308 [4] bias param range(256): [0x768f0b0, 0x768f4b0], offset: 0x9e9a0
Sun Sep  8 22:46:59 2019 D CONV ConvolutionLayer::setAddr():311 [4] preluNegtiveSlope param range(0): [0x768f4b0, 0x768f4b0], offset: 0x9eda0

使用CUDA后

fn-yang
关注
  • 2
    点赞
  • 踩
  • 5
    收藏
    觉得还不错? 一键收藏
  • 3
    评论
专栏目录
视频教程-基于海思35xx上nnie加速引擎进行yolov3模型推理-计算机视觉
weixin_33370014的博客
05-28 578
基于海思35xx上nnie加速引擎进行yolov3模型推理 2003年工学硕...
海思Hi3559AV100开发板CADENCE原理图和 PADS9.5 PCB源文件.zip
01-30
海思Hi3559AV100是一款高性能的芯片，专为视频处理和智能物联网应用设计。这款芯片集成了多种功能，包括图像信号处理器（ISP）、视频编码器、解码器以及强大的中央处理器，使其在监控摄像头、智能家居、自动驾驶等...
海思3559A/3519A查看 nnie 推理次数/ddr占用
Tourior的博客
04-24 2039
海思3559A/3519A常看 nnie 推理次数 cat /proc/umap/nnie
【新版HI3559AV100开发注意事项（四）】
WeChat_zhanggong345的博客
04-12 1053
三十三、SS928的VI信号应该是可以直接换成LVDS信号的，但是我手上的资料没有具体的更改方法，是改硬件，还是改软件？答：RTL8211FS和RTL8211F的区别就是8211F的RX/TX_DELAY是硬件设置的，而RTL8211FS的TX_DELAY是软件控制的，需要重新设置的话，否则这样能够发现网卡但是无法正常通信。另一个来源是图像处理模块设计支持的最大分辨率和总线性能，要跑一堆图像算法3DNR之类的，对于VBLANK区域大小是有要求的。你软件配置为Device模式，接入U盘肯定是没反应的。
海思hi3559AV100编译GPU驱动
阿Sir问道
05-31 3788
1、前言 GPU （Graphics Process Unit ）是3D图形处理单元，海思提供完全符合业界标准的 EGL 1.4，OpenGL ES 1.1/2.0/3.0/3.1/3.2 及OpenCL2.0 标准接口。 2、驱动编译和加载 GPU 代码路径：SDK 目录下mpp/component/gpu文件夹。GPU 驱动分内核态驱动与用户态驱动...
华为海思 AI 芯片 (Hi3559A V100) 算法开发（五） 搭建 windows caffe python 开发环境 windows10 + anacod + cuda10
随意聊点
05-18 1727
这篇文章就不讲的太过详细，因为遇到的 BUG 太多，首先网上没有 VS 2019 的版本的环境搭建，并且不向下兼容，驱动也遇到了问题，BUG 简直就是为每个人量身定做，本篇文章的环境基础是 vs2019 +windows10+ python2.7 + cuda10 + cudnn 因为翻过的文章太多，这里只放记得的一部分的环境搭建 命运就是这么颠沛流离，刚刚还失败，在这个周六加班的夜晚，突然又成功了，这里简单写一下各个环境的搭配吧，以及应该按照哪个博文进行搭配。 1. VS 版本 ...
华为海思 AI 芯片 (Hi3559A V100) 算法开发（四） 在 PC 仿真库使用 YOLOv3 进行图片目标检测以及 NMS、YOLO 讲解
随意聊点
05-06 4828
摘要 经过上一篇文章，大家编译运行 PC 仿真后，可能觉得有点迷惘，实际上你已经在后台完成了图像识别，保存结果，绘制识别框等一系列过程。这篇文章就教教大家怎么看识别的结果已经如何手动将识别框绘制出来。 仿真代码浅析 为了方便实际操作，这篇文章使用 VS 来快速运行，我觉得 vs 运行比 Ruyi 稍微快一点 为了速度再快一点，我们修改我们要运行的文件夹下...
Hi3559 ，NNIE，海思芯片AI方案
u010451780的专栏
06-17 1632
https://blog.csdn.net/zh8706/article/details/94554337 海思NNIE开发系列文章： 海思NNIE开发（一）：海思Hi3559AV100/Hi3519AV100 NNIE深度学习模块开发与调试记录 海思NNIE开发（二）：FasterRCNN在海思NNIE平台上的执行流程（一） 海思NNIE开发（三）：FasterRCNN在海思NNIE平台上的执行流程（二） 海思NNIE开发（四）:NNIE模块读入JPEG图片或视频 海思NNIE开发（五）：基于
Hi3519AV100与Hi3559AV100 开发包差异说明_hi3559av100_海思_
10-02
首先，从处理器架构来看，Hi3519AV100是一款专为低功耗、高性价比的视频监控应用设计的芯片，而Hi3559AV100则更加强调高性能和多功能性，适合于复杂场景的智能分析和处理。因此，Hi3559AV100的开发包中可能包含更...
HiMPP V4.0 媒体处理软件开发参考_hi3559av100手册_hi3559av100_
09-30
该手册首先会介绍Hi3559AV100的基本特性，包括其强大的硬件加速能力、多格式视频编解码支持、高精度图像信号处理单元以及丰富的接口资源。这些特性使得该芯片能够处理复杂的多媒体任务，实现高清视频的实时编码和...
Hi3559AV100 SDK 安装及升级使用说明.pdf
09-15
Hi3559AV100 SDK 安装及升级使用说明 本文档提供了 Hi3559AV100 SDK 的安装及升级使用说明，旨在帮助用户快速在 Hi3559AV100 DEMB 板上搭建好 SDK 运行环境。本文档适用于技术支持工程师、软件开发工程师等。 一、...
HI3559AV100dataSHEET,产品手册，Hi3559A完整介绍资料
10-28
HI3559AV100是专业的8K Ultra HDMobile Camera SOC，它提供了8K30/4K120广播级图像质量的数字视频录制，支持多路Sensor输入，支持H.265编码输出或影视级的RAW数据输出，并集成高性能ISP处理，同时采用先进低功耗工艺...
海思Hi3559AV100芯片硬件开发板评估板CADENCE（原理图+ PCB）工程文件+PDF原理图
最新发布
04-25
这些文件组合起来，构成了一个完整的硬件开发资源包，不仅适用于学习Hi3559AV100芯片的系统架构和接口设计，也适用于开发者进行二次开发或者设计相似功能的硬件产品。通过深入理解这些文件，工程师可以学习到如何将...
Hi3559AV100ES 硬件设计用户指南.pdf
03-23
Hi3559AV100ES 硬件设计用户指南.pdf 官方文件 可以参考
hi3559AV100.rar
08-07
《全面解析：Hi3559AV100平台核心技术》 Hi3559AV100是一款高性能、低功耗的视频处理芯片，广泛应用于监控摄像头、智能家居、智能交通等多个领域。此压缩包“hi3559AV100.rar”包含了关于该芯片的详细资料，包括...
Hi3559AV100ES Sensor输入接口电平场景详细说明_Hi3559AV100ESSensor_
10-01
《Hi3559AV100ES Sensor输入接口电平场景详细说明》 Hi3559AV100ES是一款高性能的芯片，广泛应用于视频处理和图像传感器控制领域。该芯片的输入接口电平设计是其核心功能之一，对于系统稳定性和数据传输效率具有至...
海思AI芯片(Hi3519A/3559A)方案学习（二十八）Ruyistudio下的仿真代码编译
ltshan139的专栏
02-23 5510
前言 目前模型仿真只能在windows上面进行。有两种方式，一是安装visual studio 2015 64bit软件，然后双击sample下面的*.sln来进入该项目，最后在VS下编译和运行或调试；二是直接在ruyistudio下编译和仿真。 这里主要讲第二种方式的项目属性设置和编译。 环境搭建 1）MinGW-W64编译工具链的安装 要在ruyistudio下运行仿真功能，必须先使用...
GPGPU-Sim：一款支持Nvidia的CUDA和OpenCL并行编程模型的仿真器
junning_wu的专栏
08-22 2105
GPGPU-Sim Welcome to the GPGPU-Sim webpage. GPGPU-Sim provides a detailed simulation model of a contemporary GPU (such as NVIDIA's Fermi and GT200 architectures) running CUDA and/or OpenCL workloads
海思3559a从零开始做计算机视觉3
qq_17278169的博客
06-11 2263
第三天 nnie移植失败了 暂时先放着吧，既然硬核跑不了，先试一下软核，使用ncnn框架来部署目标检测模型 1、交叉编译opencv Target Root选择交叉编译工具链的bin目录就行 其余的默认 然后开始编译，肯定会出错 undefined reference to `pthread_mutexattr_destroy` 修改编译目录下的 CMakeCache.txt 文件，找到 ...
hi3559av100
06-10
Hi3559AV100 是海思半导体（Hisilicon）推出的一款高性能视频处理器芯片。该芯片采用了海思自主研发的 DaVinci 架构，支持多路高清视频输入和输出，适用于广泛的视频监控、智能家居、智能交通等领域。 Hi3559AV100 芯片的主要特点如下： 1. 高性能处理：Hi3559AV100 芯片搭载了四核 Cortex-A73 和四核 Cortex-A53 处理器，能够支持多路高清视频的实时编解码和处理。 2. 多种视频输入输出接口：Hi3559AV100 芯片支持多种视频输入输出接口，包括 HDMI、CVBS、MIPI、LVDS 等，适用于不同的视频采集和输出场景。 3. 强大的图像处理能力：Hi3559AV100 芯片内置了多种图像处理算法和技术，如智能降噪、宽动态范围调节、人脸检测、车牌识别等，能够实现更加精准和智能的图像处理。 4. 多种存储接口：Hi3559AV100 芯片支持多种存储接口，包括 SATA、eMMC、SD/MMC 等，能够满足不同存储需求。 5. 低功耗设计：Hi3559AV100 芯片采用了 16nm 工艺和低功耗设计，能够实现更好的性能和能效平衡。 总之，Hi3559AV100 是一款高性能视频处理器芯片，具有强大的视频处理能力和多种视频输入输出接口，适用于广泛的视频监控、智能家居、智能交通等领域。

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论 3
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值