TensorRT6.0 parse a ssd model from caffe ,get a error. inference: nvPluginsLegacy.cpp:1026: virtual void nvinfer1::plugin::DetectionOutputLegacy::configure(const nvinfer1::Dims*, int, const nvinfer1::Dims*, int, int): Assertion `C2 == inputDims[param.in

the same code can work well in tensorRT4,here is the output when parse.

[code][GIE] retrieved output tensor ‘detection_out’
layer name:conv1_1,type:0input tensor name:data,type:03 300 300
output tensor name:(Unnamed Layer* 0) [Convolution]_output,type:064 300 300
layer name:relu1_1,type:2input tensor name:(Unnamed Layer* 0) [Convolution]_output,type:064 300 300
output tensor name:conv1_1,type:064 300 300
layer name:conv1_2,type:0input tensor name:conv1_1,type:064 300 300
output tensor name:(Unnamed Layer* 2) [Convolution]_output,type:064 300 300
layer name:relu1_2,type:2input tensor name:(Unnamed Layer* 2) [Convolution]_output,type:064 300 300
output tensor name:conv1_2,type:064 300 300
layer name:pool1,type:3input tensor name:conv1_2,type:064 300 300
output tensor name:pool1,type:064 150 150
layer name:conv2_1,type:0input tensor name:pool1,type:064 150 150
output tensor name:(Unnamed Layer* 5) [Convolution]_output,type:0128 150 150
layer name:relu2_1,type:2input tensor name:(Unnamed Layer* 5) [Convolution]_output,type:0128 150 150
output tensor name:conv2_1,type:0128 150 150
layer name:conv2_2,type:0input tensor name:conv2_1,type:0128 150 150
output tensor name:(Unnamed Layer* 7) [Convolution]_output,type:0128 150 150
layer name:relu2_2,type:2input tensor name:(Unnamed Layer* 7) [Convolution]_output,type:0128 150 150
output tensor name:conv2_2,type:0128 150 150
layer name:pool2,type:3input tensor name:conv2_2,type:0128 150 150
output tensor name:pool2,type:0128 75 75
layer name:conv3_1,type:0input tensor name:pool2,type:0128 75 75
output tensor name:(Unnamed Layer* 10) [Convolution]_output,type:0256 75 75
layer name:relu3_1,type:2input tensor name:(Unnamed Layer* 10) [Convolution]_output,type:0256 75 75
output tensor name:conv3_1,type:0256 75 75
layer name:conv3_2,type:0input tensor name:conv3_1,type:0256 75 75
output tensor name:(Unnamed Layer* 12) [Convolution]_output,type:0256 75 75
layer name:relu3_2,type:2input tensor name:(Unnamed Layer* 12) [Convolution]_output,type:0256 75 75
output tensor name:conv3_2,type:0256 75 75
layer name:conv3_3,type:0input tensor name:conv3_2,type:0256 75 75
output tensor name:(Unnamed Layer* 14) [Convolution]_output,type:0256 75 75
layer name:relu3_3,type:2input tensor name:(Unnamed Layer* 14) [Convolution]_output,type:0256 75 75
output tensor name:conv3_3,type:0256 75 75
layer name:pool3,type:3input tensor name:conv3_3,type:0256 75 75
output tensor name:pool3,type:0256 38 38
layer name:conv4_1,type:0input tensor name:pool3,type:0256 38 38
output tensor name:(Unnamed Layer* 17) [Convolution]_output,type:0512 38 38
layer name:relu4_1,type:2input tensor name:(Unnamed Layer* 17) [Convolution]_output,type:0512 38 38
output tensor name:conv4_1,type:0512 38 38
layer name:conv4_2,type:0input tensor name:conv4_1,type:0512 38 38
output tensor name:(Unnamed Layer* 19) [Convolution]_output,type:0512 38 38
layer name:relu4_2,type:2input tensor name:(Unnamed Layer* 19) [Convolution]_output,type:0512 38 38
output tensor name:conv4_2,type:0512 38 38
layer name:conv4_3,type:0input tensor name:conv4_2,type:0512 38 38
output tensor name:(Unnamed Layer* 21) [Convolution]_output,type:0512 38 38
layer name:relu4_3,type:2input tensor name:(Unnamed Layer* 21) [Convolution]_output,type:0512 38 38
output tensor name:conv4_3,type:0512 38 38
layer name:pool4,type:3input tensor name:conv4_3,type:0512 38 38
output tensor name:pool4,type:0512 19 19
layer name:conv5_1,type:0input tensor name:pool4,type:0512 19 19
output tensor name:(Unnamed Layer* 24) [Convolution]_output,type:0512 19 19
layer name:relu5_1,type:2input tensor name:(Unnamed Layer* 24) [Convolution]_output,type:0512 19 19
output tensor name:conv5_1,type:0512 19 19
layer name:conv5_2,type:0input tensor name:conv5_1,type:0512 19 19
output tensor name:(Unnamed Layer* 26) [Convolution]_output,type:0512 19 19
layer name:relu5_2,type:2input tensor name:(Unnamed Layer* 26) [Convolution]_output,type:0512 19 19
output tensor name:conv5_2,type:0512 19 19
layer name:conv5_3,type:0input tensor name:conv5_2,type:0512 19 19
output tensor name:(Unnamed Layer* 28) [Convolution]_output,type:0512 19 19
layer name:relu5_3,type:2input tensor name:(Unnamed Layer* 28) [Convolution]_output,type:0512 19 19
output tensor name:conv5_3,type:0512 19 19
layer name:pool5,type:3input tensor name:conv5_3,type:0512 19 19
output tensor name:pool5,type:0512 19 19
layer name:fc6,type:0input tensor name:pool5,type:0512 19 19
output tensor name:(Unnamed Layer* 31) [Convolution]_output,type:01024 19 19
layer name:relu6,type:2input tensor name:(Unnamed Layer* 31) [Convolution]_output,type:01024 19 19
output tensor name:fc6,type:01024 19 19
layer name:fc7,type:0input tensor name:fc6,type:01024 19 19
output tensor name:(Unnamed Layer* 33) [Convolution]_output,type:01024 19 19
layer name:relu7,type:2input tensor name:(Unnamed Layer* 33) [Convolution]_output,type:01024 19 19
output tensor name:fc7,type:01024 19 19
layer name:conv6_1,type:0input tensor name:fc7,type:01024 19 19
output tensor name:(Unnamed Layer* 35) [Convolution]_output,type:0256 19 19
layer name:conv6_1_relu,type:2input tensor name:(Unnamed Layer* 35) [Convolution]_output,type:0256 19 19
output tensor name:conv6_1,type:0256 19 19
layer name:conv6_2,type:0input tensor name:conv6_1,type:0256 19 19
output tensor name:(Unnamed Layer* 37) [Convolution]_output,type:0512 10 10
layer name:conv6_2_relu,type:2input tensor name:(Unnamed Layer* 37) [Convolution]_output,type:0512 10 10
output tensor name:conv6_2,type:0512 10 10
layer name:conv7_1,type:0input tensor name:conv6_2,type:0512 10 10
output tensor name:(Unnamed Layer* 39) [Convolution]_output,type:0128 10 10
layer name:conv7_1_relu,type:2input tensor name:(Unnamed Layer* 39) [Convolution]_output,type:0128 10 10
output tensor name:conv7_1,type:0128 10 10
layer name:conv7_2,type:0input tensor name:conv7_1,type:0128 10 10
output tensor name:(Unnamed Layer* 41) [Convolution]_output,type:0256 5 5
layer name:conv7_2_relu,type:2input tensor name:(Unnamed Layer* 41) [Convolution]_output,type:0256 5 5
output tensor name:conv7_2,type:0256 5 5
layer name:conv8_1,type:0input tensor name:conv7_2,type:0256 5 5
output tensor name:(Unnamed Layer* 43) [Convolution]_output,type:0128 5 5
layer name:conv8_1_relu,type:2input tensor name:(Unnamed Layer* 43) [Convolution]_output,type:0128 5 5
output tensor name:conv8_1,type:0128 5 5
layer name:conv8_2,type:0input tensor name:conv8_1,type:0128 5 5
output tensor name:(Unnamed Layer* 45) [Convolution]_output,type:0256 3 3
layer name:conv8_2_relu,type:2input tensor name:(Unnamed Layer* 45) [Convolution]_output,type:0256 3 3
output tensor name:conv8_2,type:0256 3 3
layer name:conv9_1,type:0input tensor name:conv8_2,type:0256 3 3
output tensor name:(Unnamed Layer* 47) [Convolution]_output,type:0128 3 3
layer name:conv9_1_relu,type:2input tensor name:(Unnamed Layer* 47) [Convolution]_output,type:0128 3 3
output tensor name:conv9_1,type:0128 3 3
layer name:conv9_2,type:0input tensor name:conv9_1,type:0128 3 3
output tensor name:(Unnamed Layer* 49) [Convolution]_output,type:0256 1 1
layer name:conv9_2_relu,type:2input tensor name:(Unnamed Layer* 49) [Convolution]_output,type:0256 1 1
output tensor name:conv9_2,type:0256 1 1
layer name:conv4_3_norm,type:10input tensor name:conv4_3,type:0512 38 38
output tensor name:conv4_3_norm,type:0512 38 38
layer name:conv4_3_norm_mbox_loc,type:0input tensor name:conv4_3_norm,type:0512 38 38
output tensor name:conv4_3_norm_mbox_loc,type:016 38 38
layer name:conv4_3_norm_mbox_loc_perm,type:10input tensor name:conv4_3_norm_mbox_loc,type:016 38 38
output tensor name:conv4_3_norm_mbox_loc_perm,type:038 38 16
layer name:conv4_3_norm_mbox_loc_flat,type:10input tensor name:conv4_3_norm_mbox_loc_perm,type:038 38 16
output tensor name:conv4_3_norm_mbox_loc_flat,type:023104 1 1
layer name:conv4_3_norm_mbox_conf,type:0input tensor name:conv4_3_norm,type:0512 38 38
output tensor name:conv4_3_norm_mbox_conf,type:040 38 38
layer name:conv4_3_norm_mbox_conf_perm,type:10input tensor name:conv4_3_norm_mbox_conf,type:040 38 38
output tensor name:conv4_3_norm_mbox_conf_perm,type:038 38 40
layer name:conv4_3_norm_mbox_conf_flat,type:10input tensor name:conv4_3_norm_mbox_conf_perm,type:038 38 40
output tensor name:conv4_3_norm_mbox_conf_flat,type:057760 1 1
layer name:conv4_3_norm_mbox_priorbox,type:10input tensor name:conv4_3_norm,type:0512 38 38
input tensor name:data,type:03 300 300
output tensor name:conv4_3_norm_mbox_priorbox,type:02 23104 1
layer name:fc7_mbox_loc,type:0input tensor name:fc7,type:01024 19 19
output tensor name:fc7_mbox_loc,type:024 19 19
layer name:fc7_mbox_loc_perm,type:10input tensor name:fc7_mbox_loc,type:024 19 19
output tensor name:fc7_mbox_loc_perm,type:019 19 24
layer name:fc7_mbox_loc_flat,type:10input tensor name:fc7_mbox_loc_perm,type:019 19 24
output tensor name:fc7_mbox_loc_flat,type:08664 1 1
layer name:fc7_mbox_conf,type:0input tensor name:fc7,type:01024 19 19
output tensor name:fc7_mbox_conf,type:060 19 19
layer name:fc7_mbox_conf_perm,type:10input tensor name:fc7_mbox_conf,type:060 19 19
output tensor name:fc7_mbox_conf_perm,type:019 19 60
layer name:fc7_mbox_conf_flat,type:10input tensor name:fc7_mbox_conf_perm,type:019 19 60
output tensor name:fc7_mbox_conf_flat,type:021660 1 1
layer name:fc7_mbox_priorbox,type:10input tensor name:fc7,type:01024 19 19
input tensor name:data,type:03 300 300
output tensor name:fc7_mbox_priorbox,type:02 8664 1
layer name:conv6_2_mbox_loc,type:0input tensor name:conv6_2,type:0512 10 10
output tensor name:conv6_2_mbox_loc,type:024 10 10
layer name:conv6_2_mbox_loc_perm,type:10input tensor name:conv6_2_mbox_loc,type:024 10 10
output tensor name:conv6_2_mbox_loc_perm,type:010 10 24
layer name:conv6_2_mbox_loc_flat,type:10input tensor name:conv6_2_mbox_loc_perm,type:010 10 24
output tensor name:conv6_2_mbox_loc_flat,type:02400 1 1
layer name:conv6_2_mbox_conf,type:0input tensor name:conv6_2,type:0512 10 10
output tensor name:conv6_2_mbox_conf,type:060 10 10
layer name:conv6_2_mbox_conf_perm,type:10input tensor name:conv6_2_mbox_conf,type:060 10 10
output tensor name:conv6_2_mbox_conf_perm,type:010 10 60
layer name:conv6_2_mbox_conf_flat,type:10input tensor name:conv6_2_mbox_conf_perm,type:010 10 60
output tensor name:conv6_2_mbox_conf_flat,type:06000 1 1
layer name:conv6_2_mbox_priorbox,type:10input tensor name:conv6_2,type:0512 10 10
input tensor name:data,type:03 300 300
output tensor name:conv6_2_mbox_priorbox,type:02 2400 1
layer name:conv7_2_mbox_loc,type:0input tensor name:conv7_2,type:0256 5 5
output tensor name:conv7_2_mbox_loc,type:024 5 5
layer name:conv7_2_mbox_loc_perm,type:10input tensor name:conv7_2_mbox_loc,type:024 5 5
output tensor name:conv7_2_mbox_loc_perm,type:05 5 24
layer name:conv7_2_mbox_loc_flat,type:10input tensor name:conv7_2_mbox_loc_perm,type:05 5 24
output tensor name:conv7_2_mbox_loc_flat,type:0600 1 1
layer name:conv7_2_mbox_conf,type:0input tensor name:conv7_2,type:0256 5 5
output tensor name:conv7_2_mbox_conf,type:060 5 5
layer name:conv7_2_mbox_conf_perm,type:10input tensor name:conv7_2_mbox_conf,type:060 5 5
output tensor name:conv7_2_mbox_conf_perm,type:05 5 60
layer name:conv7_2_mbox_conf_flat,type:10input tensor name:conv7_2_mbox_conf_perm,type:05 5 60
output tensor name:conv7_2_mbox_conf_flat,type:01500 1 1
layer name:conv7_2_mbox_priorbox,type:10input tensor name:conv7_2,type:0256 5 5
input tensor name:data,type:03 300 300
output tensor name:conv7_2_mbox_priorbox,type:02 600 1
layer name:conv8_2_mbox_loc,type:0input tensor name:conv8_2,type:0256 3 3
output tensor name:conv8_2_mbox_loc,type:016 3 3
layer name:conv8_2_mbox_loc_perm,type:10input tensor name:conv8_2_mbox_loc,type:016 3 3
output tensor name:conv8_2_mbox_loc_perm,type:03 3 16
layer name:conv8_2_mbox_loc_flat,type:10input tensor name:conv8_2_mbox_loc_perm,type:03 3 16
output tensor name:conv8_2_mbox_loc_flat,type:0144 1 1
layer name:conv8_2_mbox_conf,type:0input tensor name:conv8_2,type:0256 3 3
output tensor name:conv8_2_mbox_conf,type:040 3 3
layer name:conv8_2_mbox_conf_perm,type:10input tensor name:conv8_2_mbox_conf,type:040 3 3
output tensor name:conv8_2_mbox_conf_perm,type:03 3 40
layer name:conv8_2_mbox_conf_flat,type:10input tensor name:conv8_2_mbox_conf_perm,type:03 3 40
output tensor name:conv8_2_mbox_conf_flat,type:0360 1 1
layer name:conv8_2_mbox_priorbox,type:10input tensor name:conv8_2,type:0256 3 3
input tensor name:data,type:03 300 300
output tensor name:conv8_2_mbox_priorbox,type:02 144 1
layer name:conv9_2_mbox_loc,type:0input tensor name:conv9_2,type:0256 1 1
output tensor name:conv9_2_mbox_loc,type:016 1 1
layer name:conv9_2_mbox_loc_perm,type:10input tensor name:conv9_2_mbox_loc,type:016 1 1
output tensor name:conv9_2_mbox_loc_perm,type:01 1 16
layer name:conv9_2_mbox_loc_flat,type:10input tensor name:conv9_2_mbox_loc_perm,type:01 1 16
output tensor name:conv9_2_mbox_loc_flat,type:016 1 1
layer name:conv9_2_mbox_conf,type:0input tensor name:conv9_2,type:0256 1 1
output tensor name:conv9_2_mbox_conf,type:040 1 1
layer name:conv9_2_mbox_conf_perm,type:10input tensor name:conv9_2_mbox_conf,type:040 1 1
output tensor name:conv9_2_mbox_conf_perm,type:01 1 40
layer name:conv9_2_mbox_conf_flat,type:10input tensor name:conv9_2_mbox_conf_perm,type:01 1 40
output tensor name:conv9_2_mbox_conf_flat,type:040 1 1
layer name:conv9_2_mbox_priorbox,type:10input tensor name:conv9_2,type:0256 1 1
input tensor name:data,type:03 300 300
output tensor name:conv9_2_mbox_priorbox,type:02 16 1
layer name:mbox_loc,type:10input tensor name:conv4_3_norm_mbox_loc_flat,type:023104 1 1
input tensor name:fc7_mbox_loc_flat,type:08664 1 1
input tensor name:conv6_2_mbox_loc_flat,type:02400 1 1
input tensor name:conv7_2_mbox_loc_flat,type:0600 1 1
input tensor name:conv8_2_mbox_loc_flat,type:0144 1 1
input tensor name:conv9_2_mbox_loc_flat,type:016 1 1
output tensor name:mbox_loc,type:034928 1 1
layer name:mbox_conf,type:10input tensor name:conv4_3_norm_mbox_conf_flat,type:057760 1 1
input tensor name:fc7_mbox_conf_flat,type:021660 1 1
input tensor name:conv6_2_mbox_conf_flat,type:06000 1 1
input tensor name:conv7_2_mbox_conf_flat,type:01500 1 1
input tensor name:conv8_2_mbox_conf_flat,type:0360 1 1
input tensor name:conv9_2_mbox_conf_flat,type:040 1 1
output tensor name:mbox_conf,type:087320 1 1
layer name:mbox_priorbox,type:10input tensor name:conv4_3_norm_mbox_priorbox,type:02 23104 1
input tensor name:fc7_mbox_priorbox,type:02 8664 1
input tensor name:conv6_2_mbox_priorbox,type:02 2400 1
input tensor name:conv7_2_mbox_priorbox,type:02 600 1
input tensor name:conv8_2_mbox_priorbox,type:02 144 1
input tensor name:conv9_2_mbox_priorbox,type:02 16 1
output tensor name:mbox_priorbox,type:02 34928 1
layer name:mbox_conf_reshape,type:10input tensor name:mbox_conf,type:087320 1 1
output tensor name:mbox_conf_reshape,type:08732 10 1
layer name:mbox_conf_softmax,type:10input tensor name:mbox_conf_reshape,type:08732 10 1
output tensor name:mbox_conf_softmax,type:08732 10 1
layer name:mbox_conf_flatten,type:10input tensor name:mbox_conf_softmax,type:08732 10 1
output tensor name:mbox_conf_flatten,type:087320 1 1
layer name:detection_out,type:10input tensor name:mbox_loc,type:034928 1 1
input tensor name:mbox_conf_flatten,type:087320 1 1
input tensor name:mbox_priorbox,type:02 34928 1
output tensor name:detection_out,type:01 400 7
output tensor name:detection_out2,type:01 1 1
[GIE] configuring CUDA engine
[GIE] building CUDA engine
[GIE] Applying generic optimizations to the graph for inference.
[GIE] Original: 101 layers
[GIE] After dead-layer removal: 101 layers
[GIE] After scale fusion: 101 layers
[GIE] Fusing conv1_1 with relu1_1
[GIE] Fusing conv1_2 with relu1_2
[GIE] Fusing conv2_1 with relu2_1
[GIE] Fusing conv2_2 with relu2_2
[GIE] Fusing conv3_1 with relu3_1
[GIE] Fusing conv3_2 with relu3_2
[GIE] Fusing conv3_3 with relu3_3
[GIE] Fusing conv4_1 with relu4_1
[GIE] Fusing conv4_2 with relu4_2
[GIE] Fusing conv4_3 with relu4_3
[GIE] Fusing conv5_1 with relu5_1
[GIE] Fusing conv5_2 with relu5_2
[GIE] Fusing conv5_3 with relu5_3
[GIE] Fusing fc6 with relu6
[GIE] Fusing fc7 with relu7
[GIE] Fusing conv6_1 with conv6_1_relu
[GIE] Fusing conv6_2 with conv6_2_relu
[GIE] Fusing conv7_1 with conv7_1_relu
[GIE] Fusing conv7_2 with conv7_2_relu
[GIE] Fusing conv8_1 with conv8_1_relu
[GIE] Fusing conv8_2 with conv8_2_relu
[GIE] Fusing conv9_1 with conv9_1_relu
[GIE] Fusing conv9_2 with conv9_2_relu
[GIE] After vertical fusions: 78 layers
[GIE] After final dead-layer removal: 78 layers
[GIE] After tensor merging: 78 layers
[GIE] After concat removal: 78 layers
[GIE] Graph construction and optimization completed in 0.0104586 seconds.
[GIE] Constructing optimization profile number 0 out of 1
*************** Autotuning format combination: Float(1,300,90000,270000) → Float(1,300,90000,5760000) ***************
[GIE] --------------- Timing Runner: conv1_1 + relu1_1 (LegacySASSConvolution)
[GIE] Tactic: 0 time 7.02438
[GIE] Tactic: 1 time 14.575
[GIE] Fastest Tactic: 0 Time: 7.02438
[GIE] --------------- Timing Runner: conv1_1 + relu1_1 (FusedConvActConvolution)
[GIE] Tactic: 7 time 13.6904
[GIE] Tactic: 29 time 13.2899
[GIE] Tactic: 30 time 10.1713
[GIE] Tactic: 43 time 9.76096
[GIE] Tactic: 66 time 10.3418
[GIE] Tactic: 90 time 9.59222
[GIE] Tactic: 104 time 9.99974
[GIE] Tactic: 130 time 10.2089
[GIE] Tactic: 136 time 10.0957
[GIE] Tactic: 144 time 9.76218
[GIE] Tactic: 153 time 9.9479
[GIE] Tactic: 156 time 9.69888
[GIE] Fastest Tactic: 90 Time: 9.59222
[GIE] --------------- Timing Runner: conv1_1 + relu1_1 (CaskConvolution)
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 2.06288
[GIE] conv1_1 + relu1_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 6.8512
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 1.58371
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 3.04787
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 1.53664
[GIE] conv1_1 + relu1_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 4.5473
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 1.56704
[GIE] conv1_1 + relu1_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 6.73472
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 3.05488
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 2.10118
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 1.976
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 2.99814
[GIE] Fastest Tactic: 5137655947464784826 Time: 1.53664
[GIE] --------------- Timing Runner: conv1_1 + relu1_1 (CudaConvolution)
[GIE] Tactic: 0 time 9.47824
[GIE] Tactic: 1 time 5.82211
[GIE] Tactic: 2 time 9.13587
[GIE] Tactic: 6 time 8.7824
[GIE] Fastest Tactic: 1 Time: 5.82211
[GIE] --------------- Timing Runner: conv1_1 + relu1_1 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: CaskConvolution Tactic: 5137655947464784826
[GIE] conv1_1 + relu1_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,300,90000,5760000) → Float(1,300,90000,5760000) ***************
[GIE] --------------- Timing Runner: conv1_2 + relu1_2 (LegacySASSConvolution)
[GIE] Tactic: 0 time 15.3166
[GIE] Tactic: 1 time 9.86192
[GIE] Fastest Tactic: 1 Time: 9.86192
[GIE] --------------- Timing Runner: conv1_2 + relu1_2 (FusedConvActConvolution)
[GIE] Tactic: 7 time 18.5467
[GIE] Tactic: 10 time 17.1948
[GIE] Tactic: 14 time 18.4029
[GIE] Tactic: 15 time 18.476
[GIE] Tactic: 25 time 18.7488
[GIE] Tactic: 26 time 23.1138
[GIE] Tactic: 29 time 16.9717
[GIE] Tactic: 30 time 17.6546
[GIE] Tactic: 33 time 18.9518
[GIE] Tactic: 36 time 24.9653
[GIE] Tactic: 39 time 25.2957
[GIE] Tactic: 41 time 18.4994
[GIE] Tactic: 42 time 48.5101
[GIE] Tactic: 43 time 31.9368
[GIE] Tactic: 45 time 19.2296
[GIE] Tactic: 47 time 19.368
[GIE] Tactic: 52 time 32.1894
[GIE] Tactic: 54 time 18.8461
[GIE] Tactic: 56 time 31.7734
[GIE] Tactic: 66 time 18.0944
[GIE] Tactic: 76 time 17.979
[GIE] Tactic: 90 time 17.7424
[GIE] Tactic: 93 time 17.695
[GIE] Tactic: 98 time 18.3336
[GIE] Tactic: 104 time 18.9414
[GIE] Tactic: 110 time 24.5158
[GIE] Tactic: 119 time 28.0275
[GIE] Tactic: 121 time 17.1362
[GIE] Tactic: 130 time 17.4789
[GIE] Tactic: 134 time 23.7923
[GIE] Tactic: 136 time 18.335
[GIE] Tactic: 137 time 19.8544
[GIE] Tactic: 139 time 18.8229
[GIE] Tactic: 144 time 18.4822
[GIE] Tactic: 149 time 29.6312
[GIE] Tactic: 151 time 22.9858
[GIE] Tactic: 152 time 17.8498
[GIE] Tactic: 153 time 17.8673
[GIE] Tactic: 156 time 16.9731
[GIE] Tactic: 159 time 17.824
[GIE] Tactic: 162 time 24.1014
[GIE] Tactic: 164 time 17.4674
[GIE] Fastest Tactic: 29 Time: 16.9717
[GIE] --------------- Timing Runner: conv1_2 + relu1_2 (CaskConvolution)
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 17.3498
[GIE] conv1_2 + relu1_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 15.7653
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 14.5351
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 27.3541
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 14.0553
[GIE] conv1_2 + relu1_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 10.8686
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 14.3982
[GIE] conv1_2 + relu1_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 16.3648
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 27.6768
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 18.1438
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 16.4861
[GIE] conv1_2 + relu1_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 26.8638
[GIE] Fastest Tactic: 5921334924264294896 Time: 10.8686
[GIE] --------------- Timing Runner: conv1_2 + relu1_2 (CudaConvolution)
[GIE] Tactic: 0 time 34.9461
[GIE] Tactic: 1 time 19.708
[GIE] Tactic: 2 skipped. Scratch requested: 207360000, available: 16777216
[GIE] Tactic: 6 time 15.213
[GIE] Fastest Tactic: 6 Time: 15.213
[GIE] --------------- Timing Runner: conv1_2 + relu1_2 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,300,90000,5760000) → Float(1,150,22500,1440000) ***************
[GIE] --------------- Timing Runner: pool1 (Pooling)
[GIE] Tactic: -1 time 1.30893
[GIE] Fastest Tactic: -1 Time: 1.30893
[GIE] --------------- Timing Runner: pool1 (TiledPooling)
[GIE] Tactic: 5505281 time 2.26768
[GIE] Tactic: 5570817 time 1.57424
[GIE] Tactic: 5636353 time 1.34896
[GIE] Tactic: 5701889 time 1.33376
[GIE] Tactic: 5767425 time 1.33971
[GIE] Tactic: 5832961 time 1.34672
[GIE] Tactic: 5898497 time 1.3512
[GIE] Tactic: 5964033 time 1.38403
[GIE] Tactic: 6029569 time 2.44512
[GIE] Tactic: 6095105 time 2.22848
[GIE] Tactic: 6160641 time 2.26608
[GIE] Tactic: 6226177 time 2.26416
[GIE] Tactic: 6291713 time 2.27216
[GIE] Tactic: 6357249 time 2.27613
[GIE] Tactic: 6422785 time 2.27088
[GIE] Tactic: 6488321 time 2.25488
[GIE] Fastest Tactic: 5701889 Time: 1.33376
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: Pooling Tactic: -1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,150,22500,1440000) → Float(1,150,22500,2880000) ***************
[GIE] --------------- Timing Runner: conv2_1 + relu2_1 (LegacySASSConvolution)
[GIE] Tactic: 0 time 6.91347
[GIE] Tactic: 1 time 5.51859
[GIE] Fastest Tactic: 1 Time: 5.51859
[GIE] --------------- Timing Runner: conv2_1 + relu2_1 (FusedConvActConvolution)
[GIE] Tactic: 7 time 10.4784
[GIE] Tactic: 10 time 9.55491
[GIE] Tactic: 14 time 10.244
[GIE] Tactic: 15 time 10.9952
[GIE] Tactic: 25 time 11.7005
[GIE] Tactic: 26 time 12.9648
[GIE] Tactic: 29 time 9.00896
[GIE] Tactic: 30 time 9.49107
[GIE] Tactic: 33 time 10.4454
[GIE] Tactic: 36 time 13.6302
[GIE] Tactic: 39 time 15.1184
[GIE] Tactic: 41 time 10.1048
[GIE] Tactic: 42 time 14.6115
[GIE] Tactic: 43 time 9.64816
[GIE] Tactic: 45 time 11.3033
[GIE] Tactic: 47 time 10.7962
[GIE] Tactic: 52 time 10.3932
[GIE] Tactic: 54 time 10.7994
[GIE] Tactic: 56 time 11.1841
[GIE] Tactic: 66 time 9.76128
[GIE] Tactic: 76 time 9.47136
[GIE] Tactic: 90 time 9.53872
[GIE] Tactic: 93 time 9.7568
[GIE] Tactic: 98 time 10.7364
[GIE] Tactic: 104 time 10.7222
[GIE] Tactic: 110 time 13.7168
[GIE] Tactic: 119 time 15.693
[GIE] Tactic: 121 time 8.96307
[GIE] Tactic: 130 time 9.99219
[GIE] Tactic: 134 time 13.4772
[GIE] Tactic: 136 time 10.2866
[GIE] Tactic: 137 time 11.4832
[GIE] Tactic: 139 time 10.7237
[GIE] Tactic: 144 time 9.93056
[GIE] Tactic: 149 time 9.42979
[GIE] Tactic: 151 time 12.6104
[GIE] Tactic: 152 time 9.23376
[GIE] Tactic: 153 time 9.80928
[GIE] Tactic: 156 time 9.09424
[GIE] Tactic: 159 time 9.95904
[GIE] Tactic: 162 time 12.94
[GIE] Tactic: 164 time 9.72928
[GIE] Fastest Tactic: 121 Time: 8.96307
[GIE] --------------- Timing Runner: conv2_1 + relu2_1 (CaskConvolution)
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 8.71072
[GIE] conv2_1 + relu2_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 8.45664
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 7.27936
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 6.9672
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 7.03424
[GIE] conv2_1 + relu2_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 5.90253
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 7.20022
[GIE] conv2_1 + relu2_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 8.73872
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 7.06282
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 9.11574
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 8.24304
[GIE] conv2_1 + relu2_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 6.85325
[GIE] Fastest Tactic: 5921334924264294896 Time: 5.90253
[GIE] --------------- Timing Runner: conv2_1 + relu2_1 (CudaConvolution)
[GIE] Tactic: 0 time 14.0653
[GIE] Tactic: 1 time 9.00256
[GIE] Tactic: 2 skipped. Scratch requested: 51840000, available: 16777216
[GIE] Tactic: 6 time 8.18144
[GIE] Fastest Tactic: 6 Time: 8.18144
[GIE] --------------- Timing Runner: conv2_1 + relu2_1 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,150,22500,2880000) → Float(1,150,22500,2880000) ***************
[GIE] --------------- Timing Runner: conv2_2 + relu2_2 (LegacySASSConvolution)
[GIE] Tactic: 0 time 13.354
[GIE] Tactic: 1 time 9.70893
[GIE] Fastest Tactic: 1 Time: 9.70893
[GIE] --------------- Timing Runner: conv2_2 + relu2_2 (FusedConvActConvolution)
[GIE] Tactic: 7 time 22.4699
[GIE] Tactic: 10 time 18.2775
[GIE] Tactic: 14 time 19.8443
[GIE] Tactic: 15 time 20.9581
[GIE] Tactic: 25 time 22.5171
[GIE] Tactic: 26 time 22.562
[GIE] Tactic: 29 time 16.1603
[GIE] Tactic: 30 time 17.841
[GIE] Tactic: 33 time 18.8928
[GIE] Tactic: 36 time 24.0772
[GIE] Tactic: 39 time 26.4958
[GIE] Tactic: 41 time 18.0233
[GIE] Tactic: 42 time 26.1111
[GIE] Tactic: 43 time 17.7904
[GIE] Tactic: 45 time 20.1426
[GIE] Tactic: 47 time 19.8782
[GIE] Tactic: 52 time 19.4685
[GIE] Tactic: 54 time 18.9645
[GIE] Tactic: 56 time 22.1411
[GIE] Tactic: 66 time 17.71
[GIE] Tactic: 76 time 21.7299
[GIE] Tactic: 90 time 17.1596
[GIE] Tactic: 93 time 19.3754
[GIE] Tactic: 98 time 19.3732
[GIE] Tactic: 104 time 21.7875
[GIE] Tactic: 110 time 26.3
[GIE] Tactic: 119 time 28.7627
[GIE] Tactic: 121 time 16.3959
[GIE] Tactic: 130 time 19.1608
[GIE] Tactic: 134 time 23.5963
[GIE] Tactic: 136 time 19.0091
[GIE] Tactic: 137 time 20.8864
[GIE] Tactic: 139 time 18.9645
[GIE] Tactic: 144 time 18.2019
[GIE] Tactic: 149 time 19.2901
[GIE] Tactic: 151 time 22.0563
[GIE] Tactic: 152 time 18.7929
[GIE] Tactic: 153 time 17.6762
[GIE] Tactic: 156 time 16.1968
[GIE] Tactic: 159 time 20.1719
[GIE] Tactic: 162 time 23.1229
[GIE] Tactic: 164 time 17.761
[GIE] Fastest Tactic: 29 Time: 16.1603
[GIE] --------------- Timing Runner: conv2_2 + relu2_2 (CaskConvolution)
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 16.9176
[GIE] conv2_2 + relu2_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 15.0798
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 14.1254
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 13.509
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 13.6182
[GIE] conv2_2 + relu2_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 9.9329
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 13.974
[GIE] conv2_2 + relu2_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 15.9179
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 13.6475
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 17.5052
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 15.6385
[GIE] conv2_2 + relu2_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 13.2049
[GIE] Fastest Tactic: 5921334924264294896 Time: 9.9329
[GIE] --------------- Timing Runner: conv2_2 + relu2_2 (CudaConvolution)
[GIE] Tactic: 0 time 24.2749
[GIE] Tactic: 1 time 15.4046
[GIE] Tactic: 2 skipped. Scratch requested: 103680000, available: 16777216
[GIE] Tactic: 6 time 12.2562
[GIE] Fastest Tactic: 6 Time: 12.2562
[GIE] --------------- Timing Runner: conv2_2 + relu2_2 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,150,22500,2880000) → Float(1,75,5625,720000) ***************
[GIE] --------------- Timing Runner: pool2 (Pooling)
[GIE] Tactic: -1 time 0.646176
[GIE] Fastest Tactic: -1 Time: 0.646176
[GIE] --------------- Timing Runner: pool2 (TiledPooling)
[GIE] Tactic: 5505281 time 1.32064
[GIE] Tactic: 5570817 time 0.87296
[GIE] Tactic: 5636353 time 0.723616
[GIE] Tactic: 5701889 time 0.70416
[GIE] Tactic: 5767425 time 0.70304
[GIE] Tactic: 5832961 time 0.6912
[GIE] Tactic: 5898497 time 0.696
[GIE] Tactic: 5964033 time 0.726144
[GIE] Tactic: 6029569 time 1.16678
[GIE] Tactic: 6095105 time 0.8672
[GIE] Tactic: 6160641 time 0.90528
[GIE] Tactic: 6226177 time 0.91472
[GIE] Tactic: 6291713 time 0.9136
[GIE] Tactic: 6357249 time 0.922304
[GIE] Tactic: 6422785 time 0.91424
[GIE] Tactic: 6488321 time 0.91552
[GIE] Fastest Tactic: 5832961 Time: 0.6912
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: Pooling Tactic: -1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,75,5625,720000) → Float(1,75,5625,1440000) ***************
[GIE] --------------- Timing Runner: conv3_1 + relu3_1 (LegacySASSConvolution)
[GIE] Tactic: 0 time 6.73274
[GIE] Tactic: 1 time 5.41056
[GIE] Fastest Tactic: 1 Time: 5.41056
[GIE] --------------- Timing Runner: conv3_1 + relu3_1 (FusedConvActConvolution)
[GIE] Tactic: 7 time 13.0706
[GIE] Tactic: 10 time 10.4037
[GIE] Tactic: 14 time 11.0778
[GIE] Tactic: 15 time 11.4238
[GIE] Tactic: 25 time 12.0784
[GIE] Tactic: 26 time 11.8195
[GIE] Tactic: 29 time 9.24256
[GIE] Tactic: 30 time 10.3366
[GIE] Tactic: 33 time 10.4351
[GIE] Tactic: 36 time 12.6168
[GIE] Tactic: 39 time 13.6108
[GIE] Tactic: 41 time 9.93974
[GIE] Tactic: 42 time 13.877
[GIE] Tactic: 43 time 9.91936
[GIE] Tactic: 45 time 10.5899
[GIE] Tactic: 47 time 10.1658
[GIE] Tactic: 52 time 11.0051
[GIE] Tactic: 54 time 9.97648
[GIE] Tactic: 56 time 14.2336
[GIE] Tactic: 66 time 9.57072
[GIE] Tactic: 76 time 12.261
[GIE] Tactic: 90 time 9.82848
[GIE] Tactic: 93 time 11.4866
[GIE] Tactic: 98 time 10.3378
[GIE] Tactic: 104 time 12.6347
[GIE] Tactic: 110 time 15.5973
[GIE] Tactic: 119 time 15.0161
[GIE] Tactic: 121 time 9.71654
[GIE] Tactic: 130 time 10.7419
[GIE] Tactic: 134 time 11.6586
[GIE] Tactic: 136 time 10.635
[GIE] Tactic: 137 time 10.6594
[GIE] Tactic: 139 time 10.0488
[GIE] Tactic: 144 time 9.66624
[GIE] Tactic: 149 time 12.3178
[GIE] Tactic: 151 time 11.2896
[GIE] Tactic: 152 time 10.4863
[GIE] Tactic: 153 time 9.44384
[GIE] Tactic: 156 time 9.22954
[GIE] Tactic: 159 time 10.45
[GIE] Tactic: 162 time 12.0674
[GIE] Tactic: 164 time 9.9448
[GIE] Fastest Tactic: 156 Time: 9.22954
[GIE] --------------- Timing Runner: conv3_1 + relu3_1 (CaskConvolution)
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 8.67974
[GIE] conv3_1 + relu3_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 7.81056
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 7.09856
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 6.82736
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 6.85014
[GIE] conv3_1 + relu3_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 5.6864
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 7.03002
[GIE] conv3_1 + relu3_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 8.01008
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 6.90832
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 8.99776
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 7.90432
[GIE] conv3_1 + relu3_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 6.65696
[GIE] Fastest Tactic: 5921334924264294896 Time: 5.6864
[GIE] --------------- Timing Runner: conv3_1 + relu3_1 (CudaConvolution)
[GIE] Tactic: 0 time 11.9551
[GIE] Tactic: 1 time 7.77824
[GIE] Tactic: 2 skipped. Scratch requested: 25920000, available: 16777216
[GIE] Tactic: 6 time 6.8335
[GIE] Fastest Tactic: 6 Time: 6.8335
[GIE] --------------- Timing Runner: conv3_1 + relu3_1 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,75,5625,1440000) → Float(1,75,5625,1440000) ***************
[GIE] --------------- Timing Runner: conv3_2 + relu3_2 (LegacySASSConvolution)
[GIE] Tactic: 0 time 13.1624
[GIE] Tactic: 1 time 9.58445
[GIE] Fastest Tactic: 1 Time: 9.58445
[GIE] --------------- Timing Runner: conv3_2 + relu3_2 (FusedConvActConvolution)
[GIE] Tactic: 7 time 23.7136
[GIE] Tactic: 10 time 18.6067
[GIE] Tactic: 14 time 18.5296
[GIE] Tactic: 15 time 20.2267
[GIE] Tactic: 25 time 22.0005
[GIE] Tactic: 26 time 20.1128
[GIE] Tactic: 29 time 16.0126
[GIE] Tactic: 30 time 16.8755
[GIE] Tactic: 33 time 18.4006
[GIE] Tactic: 36 time 22.3643
[GIE] Tactic: 39 time 23.3726
[GIE] Tactic: 41 time 16.977
[GIE] Tactic: 42 time 25.0701
[GIE] Tactic: 43 time 18.5419
[GIE] Tactic: 45 time 18.5751
[GIE] Tactic: 47 time 17.5822
[GIE] Tactic: 52 time 17.4833
[GIE] Tactic: 54 time 17.8108
[GIE] Tactic: 56 time 21.7578
[GIE] Tactic: 66 time 16.8555
[GIE] Tactic: 76 time 22.655
[GIE] Tactic: 90 time 16.256
[GIE] Tactic: 93 time 21.107
[GIE] Tactic: 98 time 18.3992
[GIE] Tactic: 104 time 19.6355
[GIE] Tactic: 110 time 26.2818
[GIE] Tactic: 119 time 25.7595
[GIE] Tactic: 121 time 16.726
[GIE] Tactic: 130 time 18.472
[GIE] Tactic: 134 time 20.4654
[GIE] Tactic: 136 time 17.5298
[GIE] Tactic: 137 time 18.6938
[GIE] Tactic: 139 time 17.765
[GIE] Tactic: 144 time 17.5242
[GIE] Tactic: 149 time 22.8271
[GIE] Tactic: 151 time 19.7682
[GIE] Tactic: 152 time 17.5819
[GIE] Tactic: 153 time 16.7987
[GIE] Tactic: 156 time 16.2554
[GIE] Tactic: 159 time 19.6814
[GIE] Tactic: 162 time 21.4805
[GIE] Tactic: 164 time 17.1918
[GIE] Fastest Tactic: 29 Time: 16.0126
[GIE] --------------- Timing Runner: conv3_2 + relu3_2 (CaskConvolution)
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 17.0776
[GIE] conv3_2 + relu3_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 14.1835
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 13.993
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 13.3387
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 13.4627
[GIE] conv3_2 + relu3_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 10.0398
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 13.958
[GIE] conv3_2 + relu3_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 14.3644
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 13.5095
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 17.7238
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 15.4606
[GIE] conv3_2 + relu3_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 13.0576
[GIE] Fastest Tactic: 5921334924264294896 Time: 10.0398
[GIE] --------------- Timing Runner: conv3_2 + relu3_2 (CudaConvolution)
[GIE] Tactic: 0 time 22.3091
[GIE] Tactic: 1 time 14.2164
[GIE] Tactic: 2 skipped. Scratch requested: 51840000, available: 16777216
[GIE] Tactic: 6 time 11.7242
[GIE] Fastest Tactic: 6 Time: 11.7242
[GIE] --------------- Timing Runner: conv3_2 + relu3_2 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,75,5625,1440000) → Float(1,75,5625,1440000) ***************
[GIE] --------------- Timing Runner: conv3_3 + relu3_3 (LegacySASSConvolution)
[GIE] Tactic: 0 time 13.1718
[GIE] Tactic: 1 time 9.62051
[GIE] Fastest Tactic: 1 Time: 9.62051
[GIE] --------------- Timing Runner: conv3_3 + relu3_3 (FusedConvActConvolution)
[GIE] Tactic: 7 time 23.3781
[GIE] Tactic: 10 time 16.7795
[GIE] Tactic: 14 time 18.5201
[GIE] Tactic: 15 time 19.3736
[GIE] Tactic: 25 time 21.9488
[GIE] Tactic: 26 time 20.5129
[GIE] Tactic: 29 time 16.2182
[GIE] Tactic: 30 time 16.7177
[GIE] Tactic: 33 time 18.9584
[GIE] Tactic: 36 time 22.3997
[GIE] Tactic: 39 time 23.3274
[GIE] Tactic: 41 time 16.9748
[GIE] Tactic: 42 time 25.0295
[GIE] Tactic: 43 time 18.0021
[GIE] Tactic: 45 time 18.5318
[GIE] Tactic: 47 time 17.5638
[GIE] Tactic: 52 time 19.2565
[GIE] Tactic: 54 time 17.9326
[GIE] Tactic: 56 time 28.7213
[GIE] Tactic: 66 time 16.7182
[GIE] Tactic: 76 time 22.4235
[GIE] Tactic: 90 time 16.2957
[GIE] Tactic: 93 time 21.0395
[GIE] Tactic: 98 time 18.1593
[GIE] Tactic: 104 time 19.5933
[GIE] Tactic: 110 time 25.6465
[GIE] Tactic: 119 time 25.9084
[GIE] Tactic: 121 time 16.4353
[GIE] Tactic: 130 time 18.5748
[GIE] Tactic: 134 time 20.4191
[GIE] Tactic: 136 time 17.0885
[GIE] Tactic: 137 time 18.9419
[GIE] Tactic: 139 time 17.9381
[GIE] Tactic: 144 time 17.6781
[GIE] Tactic: 149 time 23.4393
[GIE] Tactic: 151 time 19.8302
[GIE] Tactic: 152 time 18.4853
[GIE] Tactic: 153 time 16.6773
[GIE] Tactic: 156 time 16.2226
[GIE] Tactic: 159 time 19.275
[GIE] Tactic: 162 time 21.8541
[GIE] Tactic: 164 time 17.1077
[GIE] Fastest Tactic: 29 Time: 16.2182
[GIE] --------------- Timing Runner: conv3_3 + relu3_3 (CaskConvolution)
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 16.9838
[GIE] conv3_3 + relu3_3 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 14.0472
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 14.1355
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 13.3422
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 13.4957
[GIE] conv3_3 + relu3_3 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 10.1448
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 13.8468
[GIE] conv3_3 + relu3_3 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 14.3662
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 13.5175
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 17.6786
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 15.5568
[GIE] conv3_3 + relu3_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 13.0682
[GIE] Fastest Tactic: 5921334924264294896 Time: 10.1448
[GIE] --------------- Timing Runner: conv3_3 + relu3_3 (CudaConvolution)
[GIE] Tactic: 0 time 22.3102
[GIE] Tactic: 1 time 14.2194
[GIE] Tactic: 2 skipped. Scratch requested: 51840000, available: 16777216
[GIE] Tactic: 6 time 11.3028
[GIE] Fastest Tactic: 6 Time: 11.3028
[GIE] --------------- Timing Runner: conv3_3 + relu3_3 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,75,5625,1440000) → Float(1,38,1444,369664) ***************
[GIE] --------------- Timing Runner: pool3 (Pooling)
[GIE] Tactic: -1 time 0.34016
[GIE] Fastest Tactic: -1 Time: 0.34016
[GIE] --------------- Timing Runner: pool3 (TiledPooling)
[GIE] Tactic: 5505281 time 0.8512
[GIE] Tactic: 5570817 time 0.544
[GIE] Tactic: 5636353 time 0.45808
[GIE] Tactic: 5701889 time 0.421024
[GIE] Tactic: 5767425 time 0.41648
[GIE] Tactic: 5832961 time 0.403712
[GIE] Tactic: 5898497 time 0.39888
[GIE] Tactic: 5964033 time 0.42128
[GIE] Tactic: 6029569 time 0.684
[GIE] Tactic: 6095105 time 0.50208
[GIE] Tactic: 6160641 time 0.48688
[GIE] Tactic: 6226177 time 0.47984
[GIE] Tactic: 6291713 time 0.49296
[GIE] Tactic: 6357249 time 0.495424
[GIE] Tactic: 6422785 time 0.49552
[GIE] Tactic: 6488321 time 0.497056
[GIE] Fastest Tactic: 5898497 Time: 0.39888
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: Pooling Tactic: -1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,38,1444,369664) → Float(1,38,1444,739328) ***************
[GIE] --------------- Timing Runner: conv4_1 + relu4_1 (LegacySASSConvolution)
[GIE] Tactic: 0 time 7.1304
[GIE] Tactic: 1 time 5.46848
[GIE] Fastest Tactic: 1 Time: 5.46848
[GIE] --------------- Timing Runner: conv4_1 + relu4_1 (FusedConvActConvolution)
[GIE] Tactic: 7 time 12.7336
[GIE] Tactic: 10 time 10.3442
[GIE] Tactic: 14 time 10.5163
[GIE] Tactic: 15 time 10.7477
[GIE] Tactic: 25 time 11.1538
[GIE] Tactic: 26 time 11.4981
[GIE] Tactic: 29 time 8.89184
[GIE] Tactic: 30 time 9.93795
[GIE] Tactic: 33 time 10.1363
[GIE] Tactic: 36 time 11.9094
[GIE] Tactic: 39 time 11.9275
[GIE] Tactic: 41 time 9.89264
[GIE] Tactic: 42 time 14.6989
[GIE] Tactic: 43 time 10.5708
[GIE] Tactic: 45 time 9.91488
[GIE] Tactic: 47 time 9.96214
[GIE] Tactic: 52 time 10.3823
[GIE] Tactic: 54 time 9.65232
[GIE] Tactic: 56 time 16.0707
[GIE] Tactic: 66 time 9.85933
[GIE] Tactic: 76 time 12.4933
[GIE] Tactic: 90 time 9.53766
[GIE] Tactic: 93 time 11.7069
[GIE] Tactic: 98 time 10.7286
[GIE] Tactic: 104 time 11.335
[GIE] Tactic: 110 time 14.6579
[GIE] Tactic: 119 time 14.3237
[GIE] Tactic: 121 time 8.964
[GIE] Tactic: 130 time 10.5527
[GIE] Tactic: 134 time 11.73
[GIE] Tactic: 136 time 10.2166
[GIE] Tactic: 137 time 10.853
[GIE] Tactic: 139 time 9.6312
[GIE] Tactic: 144 time 10.4682
[GIE] Tactic: 149 time 13.7469
[GIE] Tactic: 151 time 10.6565
[GIE] Tactic: 152 time 10.551
[GIE] Tactic: 153 time 9.95008
[GIE] Tactic: 156 time 8.89338
[GIE] Tactic: 159 time 10.8253
[GIE] Tactic: 162 time 11.3104
[GIE] Tactic: 164 time 9.12928
[GIE] Fastest Tactic: 29 Time: 8.89184
[GIE] --------------- Timing Runner: conv4_1 + relu4_1 (CaskConvolution)
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 9.3072
[GIE] conv4_1 + relu4_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 8.43632
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 7.67949
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 7.27568
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 7.33664
[GIE] conv4_1 + relu4_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 5.9392
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 7.62106
[GIE] conv4_1 + relu4_1 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 8.56851
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 7.3641
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 9.65546
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 8.33638
[GIE] conv4_1 + relu4_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 7.07674
[GIE] Fastest Tactic: 5921334924264294896 Time: 5.9392
[GIE] --------------- Timing Runner: conv4_1 + relu4_1 (CudaConvolution)
[GIE] Tactic: 0 time 11.5403
[GIE] Tactic: 1 time 7.66694
[GIE] Tactic: 2 time 9.53706
[GIE] Tactic: 6 time 6.90819
[GIE] Fastest Tactic: 6 Time: 6.90819
[GIE] --------------- Timing Runner: conv4_1 + relu4_1 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,38,1444,739328) → Float(1,38,1444,739328) ***************
[GIE] --------------- Timing Runner: conv4_2 + relu4_2 (LegacySASSConvolution)
[GIE] Tactic: 0 time 14.1125
[GIE] Tactic: 1 time 10.7329
[GIE] Fastest Tactic: 1 Time: 10.7329
[GIE] --------------- Timing Runner: conv4_2 + relu4_2 (FusedConvActConvolution)
[GIE] Tactic: 7 time 23.7937
[GIE] Tactic: 10 time 18.6437
[GIE] Tactic: 14 time 18.9765
[GIE] Tactic: 15 time 19.2696
[GIE] Tactic: 25 time 19.6184
[GIE] Tactic: 26 time 20.9832
[GIE] Tactic: 29 time 16.734
[GIE] Tactic: 30 time 19.201
[GIE] Tactic: 33 time 18.9774
[GIE] Tactic: 36 time 22.5517
[GIE] Tactic: 39 time 21.1909
[GIE] Tactic: 41 time 18.4093
[GIE] Tactic: 42 time 27.8804
[GIE] Tactic: 43 time 20.5502
[GIE] Tactic: 45 time 18.4894
[GIE] Tactic: 47 time 18.6165
[GIE] Tactic: 52 time 21.7851
[GIE] Tactic: 54 time 18.1145
[GIE] Tactic: 56 time 33.732
[GIE] Tactic: 66 time 18.3226
[GIE] Tactic: 76 time 23.5158
[GIE] Tactic: 90 time 18.2902
[GIE] Tactic: 93 time 20.5155
[GIE] Tactic: 98 time 19.9022
[GIE] Tactic: 104 time 23.3886
[GIE] Tactic: 110 time 26.931
[GIE] Tactic: 119 time 26.8789
[GIE] Tactic: 121 time 16.6736
[GIE] Tactic: 130 time 19.9492
[GIE] Tactic: 134 time 21.3398
[GIE] Tactic: 136 time 19.8294
[GIE] Tactic: 137 time 20.0982
[GIE] Tactic: 139 time 18.0219
[GIE] Tactic: 144 time 19.7637
[GIE] Tactic: 149 time 28.0735
[GIE] Tactic: 151 time 20.1504
[GIE] Tactic: 152 time 19.9876
[GIE] Tactic: 153 time 18.3217
[GIE] Tactic: 156 time 16.8121
[GIE] Tactic: 159 time 19.9183
[GIE] Tactic: 162 time 20.948
[GIE] Tactic: 164 time 16.7723
[GIE] Fastest Tactic: 121 Time: 16.6736
[GIE] --------------- Timing Runner: conv4_2 + relu4_2 (CaskConvolution)
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 18.4546
[GIE] conv4_2 + relu4_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 16.718
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 15.2239
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 14.4055
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 14.5308
[GIE] conv4_2 + relu4_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 11.6081
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 15.0457
[GIE] conv4_2 + relu4_2 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 16.9316
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 14.5875
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 19.1764
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 16.5309
[GIE] conv4_2 + relu4_2 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 13.9941
[GIE] Fastest Tactic: 5921334924264294896 Time: 11.6081
[GIE] --------------- Timing Runner: conv4_2 + relu4_2 (CudaConvolution)
[GIE] Tactic: 0 time 22.143
[GIE] Tactic: 1 time 14.6392
[GIE] Tactic: 2 skipped. Scratch requested: 26615808, available: 16777216
[GIE] Tactic: 6 skipped. Scratch requested: 26216448, available: 16777216
[GIE] Fastest Tactic: 1 Time: 14.6392
[GIE] --------------- Timing Runner: conv4_2 + relu4_2 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,38,1444,739328) → Float(1,38,1444,739328) ***************
[GIE] --------------- Timing Runner: conv4_3 + relu4_3 (LegacySASSConvolution)
[GIE] Tactic: 0 time 14.1092
[GIE] Tactic: 1 time 10.6181
[GIE] Fastest Tactic: 1 Time: 10.6181
[GIE] --------------- Timing Runner: conv4_3 + relu4_3 (FusedConvActConvolution)
[GIE] Tactic: 7 time 23.7923
[GIE] Tactic: 10 time 18.6427
[GIE] Tactic: 14 time 18.9711
[GIE] Tactic: 15 time 19.3427
[GIE] Tactic: 25 time 19.2848
[GIE] Tactic: 26 time 21.1512
[GIE] Tactic: 29 time 16.7761
[GIE] Tactic: 30 time 19.1845
[GIE] Tactic: 33 time 18.9262
[GIE] Tactic: 36 time 22.6162
[GIE] Tactic: 39 time 21.6156
[GIE] Tactic: 41 time 18.3481
[GIE] Tactic: 42 time 27.8821
[GIE] Tactic: 43 time 20.8936
[GIE] Tactic: 45 time 18.5627
[GIE] Tactic: 47 time 18.6619
[GIE] Tactic: 52 time 21.9635
[GIE] Tactic: 54 time 18.1163
[GIE] Tactic: 56 time 32.867
[GIE] Tactic: 66 time 18.4488
[GIE] Tactic: 76 time 24.0147
[GIE] Tactic: 90 time 18.5272
[GIE] Tactic: 93 time 19.2494
[GIE] Tactic: 98 time 19.9318
[GIE] Tactic: 104 time 22.7178
[GIE] Tactic: 110 time 27.6836
[GIE] Tactic: 119 time 27.0325
[GIE] Tactic: 121 time 16.7524
[GIE] Tactic: 130 time 19.6701
[GIE] Tactic: 134 time 21.332
[GIE] Tactic: 136 time 19.9627
[GIE] Tactic: 137 time 20.0467
[GIE] Tactic: 139 time 18.1501
[GIE] Tactic: 144 time 19.7748
[GIE] Tactic: 149 time 27.0726
[GIE] Tactic: 151 time 20.1927
[GIE] Tactic: 152 time 19.82
[GIE] Tactic: 153 time 18.2739
[GIE] Tactic: 156 time 16.8369
[GIE] Tactic: 159 time 19.892
[GIE] Tactic: 162 time 21.0686
[GIE] Tactic: 164 time 16.8979
[GIE] Fastest Tactic: 121 Time: 16.7524
[GIE] --------------- Timing Runner: conv4_3 + relu4_3 (CaskConvolution)
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 18.4687
[GIE] conv4_3 + relu4_3 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148n_nt_v1
[GIE] Tactic: 3827454225649558724 time 16.7565
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_large_nn_v1
[GIE] Tactic: 4337000649858996379 time 15.2243
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_medium_nn_v1
[GIE] Tactic: 4501471010995462441 time 14.3985
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_small_nn_v1
[GIE] Tactic: 5137655947464784826 time 14.5338
[GIE] conv4_3 + relu4_3 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148m_nt_v1
[GIE] Tactic: 5921334924264294896 time 11.5902
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x64_relu_medium_nn_v1
[GIE] Tactic: 6645123197870846056 time 15.0376
[GIE] conv4_3 + relu4_3 (scudnn_winograd) Set Tactic Name: maxwell_scudnn_winograd_128x128_ldg1_ldg4_relu_tile148t_nt_v1
[GIE] Tactic: 7852627285308570038 time 16.7469
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_large_nn_v1
[GIE] Tactic: -9137461792520977713 time 14.5707
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_large_nn_v1
[GIE] Tactic: -6092040395344634144 time 19.1876
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_small_nn_v1
[GIE] Tactic: -3456450830548107839 time 16.5262
[GIE] conv4_3 + relu4_3 (scudnn) Set Tactic Name: maxwell_scudnn_128x128_relu_small_nn_v1
[GIE] Tactic: -410470605513481746 time 14.0016
[GIE] Fastest Tactic: 5921334924264294896 Time: 11.5902
[GIE] --------------- Timing Runner: conv4_3 + relu4_3 (CudaConvolution)
[GIE] Tactic: 0 time 22.0776
[GIE] Tactic: 1 time 14.6323
[GIE] Tactic: 2 skipped. Scratch requested: 26615808, available: 16777216
[GIE] Tactic: 6 skipped. Scratch requested: 26216448, available: 16777216
[GIE] Fastest Tactic: 1 Time: 14.6323
[GIE] --------------- Timing Runner: conv4_3 + relu4_3 (CudaDepthwiseConvolution)
[GIE] CudaDepthwiseConvolution has no valid tactics for this config, skipping
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: LegacySASSConvolution Tactic: 1
[GIE]
[GIE] *************** Autotuning format combination: Float(1,38,1444,739328) → Float(1,19,361,184832) ***************
[GIE] --------------- Timing Runner: pool4 (Pooling)
[GIE] Tactic: -1 time 0.172
[GIE] Fastest Tactic: -1 Time: 0.172
[GIE] --------------- Timing Runner: pool4 (TiledPooling)
[GIE] Tactic: 5505281 time 0.431232
[GIE] Tactic: 5570817 time 0.26704
[GIE] Tactic: 5636353 time 0.205792
[GIE] Tactic: 5701889 time 0.18112
[GIE] Tactic: 5767425 time 0.16672
[GIE] Tactic: 5832961 time 0.159712
[GIE] Tactic: 5898497 time 0.157376
[GIE] Tactic: 5964033 time 0.172
[GIE] Tactic: 6029569 time 0.43888
[GIE] Tactic: 6095105 time 0.288992
[GIE] Tactic: 6160641 time 0.248416
[GIE] Tactic: 6226177 time 0.239712
[GIE] Tactic: 6291713 time 0.24
[GIE] Tactic: 6357249 time 0.24928
[GIE] Tactic: 6422785 time 0.245056
[GIE] Tactic: 6488321 time 0.248704
[GIE] Fastest Tactic: 5898497 Time: 0.157376
[GIE] >>>>>>>>>>>>>>> Chose Runner Type: TiledPooling Tactic: 5898497
[GIE]
[GIE] *************** Autotuning format combination: Float(1,19,361,184832) → Float(1,19,361,184832) ***************
[GIE] --------------- Timing Runner: conv5_1 + relu5_1 (LegacySASSConvolution)
[GIE] Tactic: 0 time 3.54352
[GIE] Tactic: 1 time 4.07008
[GIE] Fastest Tactic: 0 Time: 3.54352
[GIE] --------------- Timing Runner: conv5_1 + relu5_1 (FusedConvActConvolution)
[GIE] Tactic: 7 time 4.92269
[GIE] Tactic: 10 time 5.08282
[GIE] Tactic: 14 time 4.6129
[GIE] Tactic: 15 time 5.328
[GIE] Tactic: 25 time 5.28096
[GIE] Tactic: 26 time 5.17706
[GIE] Tactic: 29 time 4.8951
[GIE] Tactic: 30 time 4.712
[GIE] Tactic: 33 time 6.24838
[GIE] Tactic: 36 time 7.1704
[GIE] Tactic: 39 time 6.04064
[GIE] Tactic: 41 time 4.47603
[GIE] Tactic: 42 time 7.00461
[GIE] Tactic: 43 time 4.7304
[GIE] Tactic: 45 time 5.41856
[GIE] Tactic: 47 time 4.55046
[GIE] Tactic: 52 time 4.82605
[GIE] Tactic: 54 time 5.2953
[GIE] Tactic: 56 time 6.89174
[GIE] Tactic: 66 time 4.37856
[GIE] Tactic: 76 time 5.38656
[GIE] Tactic: 90 time 4.51392
[GIE] Tactic: 93 time 4.65478
[GIE] Tactic: 98 time 4.90128
[GIE] Tactic: 104 time 5.36077
[GIE] Tactic: 110 time 7.20554
[GIE] Tactic: 119 time 7.46384
[GIE] Tactic: 121 time 4.89315
[GIE] Tactic: 130 time 4.6128
[GIE] Tactic: 134 time 5.17155
[GIE] Tactic: 136 time 4.67632
[GIE] Tactic: 137 time 4.89939
[GIE] Tactic: 139 time 5.30064
[GIE] Tactic: 144 time 4.7232
[GIE] Tactic: 149 time 6.70704
[GIE] Tactic: 151 time 5.696
[GIE] Tactic: 152 time 4.73776
[GIE] Tactic: 153 time 4.3672
[GIE] Tactic: 156 time 4.89718
[GIE] Tactic: 159 time 4.74525
[GIE] Tactic: 162 time 5.97238
[GIE] Tactic: 164 time 4.88077
[GIE] Fastest Tactic: 153 Time: 4.3672
[GIE] --------------- Timing Runner: conv5_1 + relu5_1 (CaskConvolution)
[GIE] conv5_1 + relu5_1 (scudnn) Set Tactic Name: maxwell_scudnn_128x32_relu_medium_nn_v1
[GIE] Tactic: 1062367460111450758 time 4.69088
[GIE] conv5_1 + relu5_1 (scudnn_winograd) Set T

Hi,

Could you please provide details on the platforms you are using:
o Linux distro and version
o GPU type
o Nvidia driver version
o CUDA version
o CUDNN version
o Python version [if using python]
o Tensorflow and PyTorch version
o TensorRT version
If possible, please share the script & model file to reproduce the issue.

Thanks