nnicolosi/Raptor
1
0
Fork 0
Code Issues Pull Requests Actions 9 Packages Projects Releases Wiki Activity

New Operations to test

Browse Source
This commit is contained in:
ilgeco
2026-08-06 14:33:42 +02:00
parent aeedf2f566
commit 0712c5ba29
10 changed files with 177 additions and 168 deletions
Download Patch File Download Diff File Expand all files Collapse all files
validation/operations/conv/conv_1x3x224x224_w64x3x7x7_b64/conv_1x3x224x224_w64x3x7x7_b64.onnx
BIN
View File
Binary file not shown.
validation/operations/conv/output_channel_grouping_minimal/c64x256.onnx
BIN
View File
Binary file not shown.
validation/operations/conv/strategy_depthwise_16/conv_strategy_depthwise_16.onnx
BIN
View File
Binary file not shown.
validation/operations/conv/strategy_input_k_tiled/conv_strategy_input_k_tiled.onnx
BIN
View File
Binary file not shown.
validation/operations/conv/strategy_output_channel_tiled/conv_strategy_output_channel_tiled.onnx
BIN
View File
Binary file not shown.
validation/operations/conv/strategy_streamed_packed/conv_strategy_streamed_packed.onnx
BIN
View File
Binary file not shown.
validation/operations/conv/strategy_streamed_patch/conv_strategy_streamed_patch.onnx
BIN
View File
Binary file not shown.
validation/operations/conv/strategy_tiled_2d/conv_strategy_tiled_2d.onnx
BIN
View File
Binary file not shown.
validation/operations/mul/after_conv_scalar_constant/mul_after_conv_scalar_constant.onnx
BIN
View File
Binary file not shown.
validation/operations/validation_results.csv
+177 -168
View File
@@ -1,169 +1,178 @@
Operation,Result,Compile,Host mem,Cores mem,Cores,Xbars,Latency,Power,Energy Operation,Result,Compile,Host mem,Cores mem,Cores,Xbars,Latency,Power,Energy
add/after_gemm,PASS,0.063 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP add/after_gemm,PASS,-,0.01 MiB,0.01 MiB,5,4,0.007784 ms,104.703618 mW,815012.960000 pJ
add/basic,PASS,0.057 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP add/basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
add/broadcast_row,PASS,0.057 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP add/broadcast_row,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
add/channel_broadcast_1024,PASS,0.061 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP add/channel_broadcast_1024,PASS,-,0.02 MiB,0.01 MiB,1,0,0.006913 ms,78.118038 mW,540030.000000 pJ
add/leading_dimension_broadcast,PASS,0.057 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP add/leading_dimension_broadcast,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
concat/channel_axis,PASS,0.069 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP concat/channel_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000457 ms,78.157549 mW,35718.000000 pJ
concat/negative_axis,PASS,0.061 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP concat/negative_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,0.001043 ms,78.092042 mW,81450.000000 pJ
concat/three_inputs_channel_axis,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP concat/three_inputs_channel_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000644 ms,78.149068 mW,50328.000000 pJ
conv/batch_2,PASS,0.066 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP conv/batch_2,PASS,-,0.00 MiB,0.00 MiB,2,2,0.013694 ms,82.623885 mW,1131451.480000 pJ
conv/batch_4_pointwise,PASS,0.065 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP conv/batch_4_pointwise,PASS,-,0.00 MiB,0.01 MiB,5,4,0.003932 ms,116.078576 mW,456420.960000 pJ
conv/depthwise_1024_channels,PASS,0.083 s,0.19 MiB,0.38 MiB,129,128,SKIP,SKIP,SKIP conv/conv_1x3x224x224_w64x3x7x7_b64,PASS,-,24.14 MiB,61.87 MiB,168,169,38.414551 ms,185.256473 mW,7116544212.120002 pJ
conv/depthwise_grouped,PASS,0.071 s,0.01 MiB,0.00 MiB,5,4,SKIP,SKIP,SKIP conv/depthwise_1024_channels,PASS,-,0.19 MiB,0.38 MiB,129,128,0.220751 ms,178.454307 mW,39393966.720000 pJ
conv/dilated_3x3,PASS,0.068 s,0.01 MiB,0.01 MiB,10,9,SKIP,SKIP,SKIP conv/depthwise_grouped,PASS,-,0.01 MiB,0.00 MiB,5,4,0.006234 ms,107.776541 mW,671878.960000 pJ
conv/dynamic,PASS,0.065 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP conv/dilated_3x3,PASS,-,0.01 MiB,0.01 MiB,10,9,0.008713 ms,118.767263 mW,1034819.160000 pJ
conv/explicit_padding,PASS,0.062 s,0.01 MiB,0.02 MiB,17,16,SKIP,SKIP,SKIP conv/dynamic,PASS,-,0.00 MiB,0.00 MiB,5,0,0.001835 ms,92.281199 mW,169336.000000 pJ
conv/grouped_many_groups,PASS,0.442 s,0.05 MiB,0.09 MiB,65,64,SKIP,SKIP,SKIP conv/explicit_padding,PASS,-,0.01 MiB,0.02 MiB,17,16,0.010007 ms,145.338047 mW,1454397.840000 pJ
conv/grouped_two_groups,PASS,0.060 s,0.00 MiB,0.00 MiB,3,2,SKIP,SKIP,SKIP conv/grouped_many_groups,PASS,-,0.05 MiB,0.09 MiB,65,64,0.181897 ms,142.207471 mW,25867112.360000 pJ
conv/huge_pointwise_1024,PASS,0.156 s,0.01 MiB,0.11 MiB,73,64,SKIP,SKIP,SKIP conv/grouped_two_groups,PASS,-,0.00 MiB,0.00 MiB,3,2,0.005361 ms,101.457653 mW,543914.480000 pJ
conv/huge_pointwise_1024_dynamic,PASS,0.079 s,8.04 MiB,12.61 MiB,168,0,SKIP,SKIP,SKIP conv/huge_pointwise_1024,PASS,-,0.01 MiB,0.11 MiB,73,64,0.015615 ms,249.545140 mW,3896647.360000 pJ
conv/kernel_3x3,PASS,0.062 s,0.01 MiB,0.01 MiB,10,9,SKIP,SKIP,SKIP conv/huge_pointwise_1024_dynamic,PASS,-,8.04 MiB,12.61 MiB,168,0,2.627964 ms,169.518697 mW,445489032.000000 pJ
conv/kernel_equals_input_spatial,PASS,0.064 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP conv/kernel_3x3,PASS,-,0.01 MiB,0.01 MiB,10,9,0.007186 ms,123.801859 mW,889640.160000 pJ
conv/large_input_channels_1x1,PASS,0.089 s,0.01 MiB,0.02 MiB,9,8,SKIP,SKIP,SKIP conv/kernel_equals_input_spatial,PASS,-,0.00 MiB,0.00 MiB,2,2,0.004639 ms,89.607562 mW,415689.480000 pJ
conv/large_output_channels_1x1,PASS,0.098 s,0.01 MiB,0.02 MiB,17,8,SKIP,SKIP,SKIP conv/large_input_channels_1x1,PASS,-,0.01 MiB,0.02 MiB,9,8,0.007648 ms,117.824519 mW,901121.920000 pJ
conv/large_spatial,PASS,0.069 s,0.01 MiB,0.04 MiB,37,36,SKIP,SKIP,SKIP conv/large_output_channels_1x1,PASS,-,0.01 MiB,0.02 MiB,17,8,0.008871 ms,128.442782 mW,1139415.920000 pJ
conv/multi_channel,PASS,0.066 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP conv/large_spatial,PASS,-,0.01 MiB,0.04 MiB,37,36,0.017018 ms,172.073372 mW,2928344.640000 pJ
conv/non_square_kernel_1x3,PASS,0.061 s,0.00 MiB,0.00 MiB,3,2,SKIP,SKIP,SKIP conv/multi_channel,PASS,-,0.00 MiB,0.00 MiB,4,3,0.006482 ms,105.683542 mW,685040.720000 pJ
conv/non_square_kernel_3x1,PASS,0.064 s,0.00 MiB,0.00 MiB,3,2,SKIP,SKIP,SKIP conv/non_square_kernel_1x3,PASS,-,0.00 MiB,0.00 MiB,3,2,0.006842 ms,99.349968 mW,679752.480000 pJ
conv/non_uniform_stride,PASS,0.062 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP conv/non_square_kernel_3x1,PASS,-,0.00 MiB,0.00 MiB,3,2,0.013484 ms,95.889683 mW,1292976.480000 pJ
conv/pointwise_1x1,PASS,0.059 s,0.00 MiB,0.00 MiB,1,1,SKIP,SKIP,SKIP conv/non_uniform_stride,PASS,-,0.00 MiB,0.00 MiB,4,3,0.007601 ms,104.048772 mW,790874.720000 pJ
conv/pointwise_tiled_chain,PASS,0.604 s,0.01 MiB,0.04 MiB,20,80,SKIP,SKIP,SKIP conv/output_channel_grouping_minimal,PASS,-,0.10 MiB,0.34 MiB,131,128,0.258453 ms,170.730913 mW,44125916.720000 pJ
conv/real_asymmetric_padding,PASS,0.060 s,0.01 MiB,0.03 MiB,29,28,SKIP,SKIP,SKIP conv/pointwise_1x1,PASS,-,0.00 MiB,0.00 MiB,1,1,0.012303 ms,80.244188 mW,987244.240000 pJ
conv/relu_conv_store,PASS,0.091 s,0.16 MiB,0.67 MiB,168,184,SKIP,SKIP,SKIP conv/pointwise_tiled_chain,PASS,-,0.01 MiB,0.04 MiB,20,80,0.041886 ms,153.880896 mW,6445455.200000 pJ
conv/same_lower_3x3,PASS,0.078 s,0.01 MiB,0.02 MiB,26,25,SKIP,SKIP,SKIP conv/real_asymmetric_padding,PASS,-,0.01 MiB,0.03 MiB,29,28,0.014457 ms,153.669968 mW,2221606.720000 pJ
conv/same_padding_3x3,PASS,0.070 s,0.01 MiB,0.02 MiB,26,25,SKIP,SKIP,SKIP conv/relu_conv_store,PASS,-,0.16 MiB,0.67 MiB,168,184,0.562898 ms,183.084489 mW,103057892.800000 pJ
conv/simple,PASS,0.064 s,0.00 MiB,0.00 MiB,1,1,SKIP,SKIP,SKIP conv/same_lower_3x3,PASS,-,0.01 MiB,0.02 MiB,26,25,0.013331 ms,166.154752 mW,2215009.000000 pJ
conv/stride_2,PASS,0.064 s,0.01 MiB,0.00 MiB,5,4,SKIP,SKIP,SKIP conv/same_padding_3x3,PASS,-,0.01 MiB,0.02 MiB,26,25,0.013331 ms,166.154752 mW,2215009.000000 pJ
conv/with_bias_3x3,PASS,0.067 s,0.00 MiB,0.01 MiB,4,3,SKIP,SKIP,SKIP conv/simple,PASS,-,0.00 MiB,0.00 MiB,1,1,0.004301 ms,83.833583 mW,360568.240000 pJ
conv/with_constant,PASS,0.070 s,0.00 MiB,0.00 MiB,1,1,SKIP,SKIP,SKIP conv/strategy_depthwise_16,PASS,-,0.06 MiB,0.35 MiB,168,168,0.335115 ms,197.936467 mW,66331479.080000 pJ
conv/without_kernel_shape_attr,PASS,0.069 s,0.01 MiB,0.01 MiB,10,9,SKIP,SKIP,SKIP conv/strategy_input_k_tiled,PASS,-,0.08 MiB,0.27 MiB,109,108,0.353736 ms,170.812713 mW,60422605.920000 pJ
conv/yolo11n_depthwise_head,PASS,1.482 s,8.66 MiB,34.24 MiB,168,255,SKIP,SKIP,SKIP conv/strategy_output_channel_tiled,PASS,-,0.03 MiB,0.16 MiB,74,72,0.091463 ms,155.743189 mW,14244739.280000 pJ
conv/yolo11n_heavy,PASS,0.431 s,4.82 MiB,19.10 MiB,161,800,SKIP,SKIP,SKIP conv/strategy_streamed_packed,PASS,-,3.34 MiB,7.89 MiB,168,168,9.353833 ms,179.858301 mW,1682364509.560000 pJ
conv/yolo11n_stem,PASS,0.783 s,12.86 MiB,37.59 MiB,168,488,SKIP,SKIP,SKIP conv/strategy_streamed_patch,PASS,-,0.34 MiB,1.32 MiB,168,168,1.904655 ms,181.910449 mW,346476645.640000 pJ
div/after_gemm,PASS,0.072 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP conv/strategy_tiled_2d,PASS,-,0.11 MiB,0.44 MiB,168,168,0.415594 ms,182.127047 mW,75690907.840000 pJ
div/basic,PASS,0.061 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP conv/stride_2,PASS,-,0.01 MiB,0.00 MiB,5,4,0.005237 ms,110.780019 mW,580154.960000 pJ
div/channel_broadcast_1024,PASS,0.066 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP conv/with_bias_3x3,PASS,-,0.00 MiB,0.01 MiB,4,3,0.007452 ms,104.162738 mW,776220.720000 pJ
div/leading_dimension_broadcast,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP conv/with_constant,PASS,-,0.00 MiB,0.00 MiB,1,1,0.006622 ms,81.738182 mW,541270.240000 pJ
div/runtime_scalar_rhs,PASS,0.056 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP conv/without_kernel_shape_attr,PASS,-,0.01 MiB,0.01 MiB,10,9,0.007186 ms,123.801859 mW,889640.160000 pJ
div/scalar_constant,PASS,0.073 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP conv/yolo11n_depthwise_head,PASS,-,8.66 MiB,34.24 MiB,168,255,42.701404 ms,200.519161 mW,8562449708.000010 pJ
gather/3d_input_axis1,PASS,0.060 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP conv/yolo11n_heavy,PASS,-,4.82 MiB,19.10 MiB,161,800,8.535404 ms,350.863768 mW,2994764012.000010 pJ
gather/axis0_matrix_indices,PASS,0.056 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP conv/yolo11n_stem,PASS,-,12.86 MiB,37.59 MiB,168,488,14.239985 ms,301.233376 mW,4289558753.000010 pJ
gather/axis1,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP div/after_gemm,PASS,-,0.01 MiB,0.01 MiB,5,4,0.007784 ms,104.703618 mW,815012.960000 pJ
gather/negative_axis,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP div/basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
gather/negative_indices,PASS,0.063 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP div/channel_broadcast_1024,PASS,-,0.02 MiB,0.01 MiB,1,0,0.006913 ms,78.118038 mW,540030.000000 pJ
gemm/alpha_beta,PASS,0.067 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP div/leading_dimension_broadcast,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
gemm/bias_rank2_broadcast,PASS,0.060 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP div/runtime_scalar_rhs,PASS,-,0.02 MiB,0.01 MiB,1,0,0.006913 ms,78.118038 mW,540030.000000 pJ
gemm/dynamic,PASS,0.064 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP div/scalar_constant,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
gemm/dynamic_alpha,PASS,0.062 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP gather/3d_input_axis1,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000589 ms,78.081494 mW,45990.000000 pJ
gemm/dynamic_beta,PASS,0.060 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP gather/axis0_matrix_indices,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000697 ms,78.068867 mW,54414.000000 pJ
gemm/dynamic_bias,PASS,0.058 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP gather/axis1,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000801 ms,78.059925 mW,62526.000000 pJ
gemm/dynamic_bias_alpha_beta,PASS,0.067 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP gather/negative_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,0.001437 ms,78.033403 mW,112134.000000 pJ
gemm/dynamic_transB,PASS,0.062 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP gather/negative_indices,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000376 ms,78.127660 mW,29376.000000 pJ
gemm/huge_1024,PASS,0.147 s,0.01 MiB,0.10 MiB,73,64,SKIP,SKIP,SKIP gemm/alpha_beta,PASS,-,0.01 MiB,0.01 MiB,5,4,0.007456 ms,105.272125 mW,784908.960000 pJ
gemm/large,PASS,0.068 s,0.02 MiB,0.03 MiB,17,16,SKIP,SKIP,SKIP gemm/bias_rank2_broadcast,PASS,-,0.00 MiB,0.01 MiB,5,4,0.007072 ms,105.979208 mW,749484.960000 pJ
gemm/large_k_small_n,PASS,0.095 s,0.01 MiB,0.01 MiB,9,8,SKIP,SKIP,SKIP gemm/dynamic,PASS,-,0.00 MiB,0.00 MiB,5,0,0.002421 ms,91.480793 mW,221475.000000 pJ
gemm/non_square,PASS,0.063 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemm/dynamic_alpha,PASS,-,0.00 MiB,0.00 MiB,5,0,0.003262 ms,91.415696 mW,298198.000000 pJ
gemm/scalar_bias,PASS,0.060 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemm/dynamic_beta,PASS,-,0.00 MiB,0.00 MiB,5,0,0.004365 ms,91.316151 mW,398595.000000 pJ
gemm/simple,PASS,0.072 s,0.03 MiB,0.08 MiB,42,40,SKIP,SKIP,SKIP gemm/dynamic_bias,PASS,-,0.00 MiB,0.00 MiB,5,0,0.002665 ms,91.445779 mW,243703.000000 pJ
gemm/small,PASS,0.065 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP gemm/dynamic_bias_alpha_beta,PASS,-,0.00 MiB,0.00 MiB,5,0,0.005629 ms,91.279268 mW,513811.000000 pJ
gemm/small_k_large_n,PASS,0.097 s,0.01 MiB,0.02 MiB,17,8,SKIP,SKIP,SKIP gemm/dynamic_transB,PASS,-,0.00 MiB,0.00 MiB,5,0,0.001301 ms,91.378171 mW,118883.000000 pJ
gemm/transA,PASS,0.064 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemm/huge_1024,PASS,-,0.01 MiB,0.10 MiB,73,64,0.017522 ms,215.037402 mW,3767885.360000 pJ
gemm/transA_transB,PASS,0.069 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemm/large,PASS,-,0.02 MiB,0.03 MiB,17,16,0.011229 ms,140.152181 mW,1573768.840000 pJ
gemm/transB,PASS,0.062 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemm/large_k_small_n,PASS,-,0.01 MiB,0.01 MiB,9,8,0.004748 ms,133.481449 mW,633769.920000 pJ
gemm/transB_with_bias,PASS,0.055 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemm/non_square,PASS,-,0.00 MiB,0.01 MiB,5,4,0.003527 ms,118.958310 mW,419565.960000 pJ
gemm/with_bias,PASS,0.067 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemm/scalar_bias,PASS,-,0.00 MiB,0.01 MiB,5,4,0.007072 ms,105.979208 mW,749484.960000 pJ
gemv/constant,PASS,0.064 s,0.00 MiB,0.00 MiB,0,0,SKIP,SKIP,SKIP gemm/simple,PASS,-,0.03 MiB,0.08 MiB,42,40,0.021640 ms,151.774196 mW,3284393.600000 pJ
gemv/simple,PASS,0.069 s,0.00 MiB,0.01 MiB,6,4,SKIP,SKIP,SKIP gemm/small,PASS,-,0.00 MiB,0.00 MiB,2,2,0.004420 ms,90.144000 mW,398436.480000 pJ
gemv/with_heterogeneous_constant,PASS,0.066 s,0.00 MiB,0.01 MiB,6,4,SKIP,SKIP,SKIP gemm/small_k_large_n,PASS,-,0.01 MiB,0.02 MiB,17,8,0.007962 ms,131.005014 mW,1043061.920000 pJ
gemv/with_homogeneous_constant,PASS,0.070 s,0.00 MiB,0.01 MiB,6,4,SKIP,SKIP,SKIP gemm/transA,PASS,-,0.00 MiB,0.01 MiB,5,4,0.005762 ms,109.140743 mW,628868.960000 pJ
gemv/with_scalar_constant,PASS,0.070 s,0.00 MiB,0.01 MiB,6,4,SKIP,SKIP,SKIP gemm/transA_transB,PASS,-,0.00 MiB,0.01 MiB,5,4,0.005762 ms,109.140743 mW,628868.960000 pJ
matmul/basic,PASS,0.062 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP gemm/transB,PASS,-,0.00 MiB,0.01 MiB,5,4,0.003527 ms,118.958310 mW,419565.960000 pJ
matmul/batched_3d,PASS,0.066 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemm/transB_with_bias,PASS,-,0.01 MiB,0.01 MiB,5,4,0.005046 ms,110.546762 mW,557818.960000 pJ
matmul/batched_3d_dynamic,PASS,0.057 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP gemm/with_bias,PASS,-,0.01 MiB,0.01 MiB,5,4,0.005562 ms,108.767882 mW,604966.960000 pJ
matmul/batched_left_constant,PASS,0.070 s,0.00 MiB,0.02 MiB,9,8,SKIP,SKIP,SKIP gemv/constant,PASS,-,0.00 MiB,0.00 MiB,0,0,0.000000 ms,2.000000 mW,0.000000 pJ
matmul/batched_lhs_broadcast,PASS,0.063 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemv/simple,PASS,-,0.00 MiB,0.01 MiB,6,4,0.005160 ms,111.150380 mW,573535.960000 pJ
matmul/batched_rhs_broadcast,PASS,0.062 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP gemv/with_heterogeneous_constant,PASS,-,0.00 MiB,0.01 MiB,6,4,0.005549 ms,109.816536 mW,609371.960000 pJ
matmul/dynamic,PASS,0.060 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP gemv/with_homogeneous_constant,PASS,-,0.00 MiB,0.01 MiB,6,4,0.005549 ms,109.816536 mW,609371.960000 pJ
matmul/huge_1024,PASS,0.145 s,0.01 MiB,0.10 MiB,73,64,SKIP,SKIP,SKIP gemv/with_scalar_constant,PASS,-,0.00 MiB,0.01 MiB,6,4,0.005549 ms,109.816536 mW,609371.960000 pJ
matmul/left_constant,PASS,0.051 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP matmul/basic,PASS,-,0.00 MiB,0.00 MiB,2,2,0.004420 ms,90.144000 mW,398436.480000 pJ
matmul/matrix_vector,PASS,0.095 s,0.52 MiB,0.78 MiB,168,173,SKIP,SKIP,SKIP matmul/batched_3d,PASS,-,0.00 MiB,0.01 MiB,5,4,0.005958 ms,108.588949 mW,646972.960000 pJ
matmul/vector_matrix,PASS,0.087 s,0.01 MiB,0.01 MiB,9,8,SKIP,SKIP,SKIP matmul/batched_3d_dynamic,PASS,-,0.00 MiB,0.00 MiB,4,0,0.001822 ms,92.192645 mW,167975.000000 pJ
matmul/yolo_attention,PASS,0.385 s,1.02 MiB,43.44 MiB,168,0,SKIP,SKIP,SKIP matmul/batched_left_constant,PASS,-,0.00 MiB,0.02 MiB,9,8,0.008822 ms,114.385164 mW,1009105.920000 pJ
mul/after_conv,PASS,0.055 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP matmul/batched_lhs_broadcast,PASS,-,0.00 MiB,0.01 MiB,5,4,0.005681 ms,109.389361 mW,621440.960000 pJ
mul/basic,PASS,0.055 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP matmul/batched_rhs_broadcast,PASS,-,0.00 MiB,0.01 MiB,5,4,0.005958 ms,108.588949 mW,646972.960000 pJ
mul/channel_broadcast_1024,PASS,0.058 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP matmul/dynamic,PASS,-,0.00 MiB,0.00 MiB,5,0,0.001621 ms,91.421962 mW,148195.000000 pJ
mul/leading_dimension_broadcast,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP matmul/huge_1024,PASS,-,0.01 MiB,0.10 MiB,73,64,0.017522 ms,215.037402 mW,3767885.360000 pJ
mul/scalar_constant,PASS,0.056 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP matmul/left_constant,PASS,-,0.00 MiB,0.01 MiB,5,4,0.005853 ms,108.861944 mW,637168.960000 pJ
pool/avg_basic,PASS,0.063 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP matmul/matrix_vector,PASS,-,0.52 MiB,0.78 MiB,168,173,0.384660 ms,202.131271 mW,77751814.880000 pJ
pool/avg_ceil_mode,PASS,0.063 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP matmul/vector_matrix,PASS,-,0.01 MiB,0.01 MiB,9,8,0.007409 ms,118.680243 mW,879301.920000 pJ
pool/avg_explicit_padding,PASS,0.061 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP matmul/yolo_attention,PASS,-,1.02 MiB,43.44 MiB,168,0,8.151445 ms,170.003707 mW,1385775865.000000 pJ
pool/avg_include_pad,PASS,0.049 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP mul/after_conv,PASS,-,0.00 MiB,0.00 MiB,4,3,0.005453 ms,107.639046 mW,586955.720000 pJ
pool/avg_large_channels,PASS,0.059 s,0.04 MiB,0.02 MiB,1,0,SKIP,SKIP,SKIP mul/after_conv_scalar_constant,PASS,-,0.00 MiB,0.00 MiB,4,3,0.005453 ms,107.639046 mW,586955.720000 pJ
pool/avg_non_uniform_stride,PASS,0.066 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP mul/basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
pool/avg_real_asymmetric_padding,PASS,0.055 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP mul/channel_broadcast_1024,PASS,-,0.02 MiB,0.01 MiB,1,0,0.006913 ms,78.118038 mW,540030.000000 pJ
pool/max_after_conv,PASS,0.069 s,0.00 MiB,0.00 MiB,5,4,SKIP,SKIP,SKIP mul/leading_dimension_broadcast,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
pool/max_basic,PASS,0.064 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP mul/scalar_constant,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
pool/max_ceil_mode,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP pool/avg_basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.011939 ms,78.022112 mW,931506.000000 pJ
pool/max_global_style_kernel_equals_input,PASS,0.063 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP pool/avg_ceil_mode,PASS,-,0.00 MiB,0.00 MiB,1,0,0.004359 ms,78.033035 mW,340146.000000 pJ
pool/max_non_square_kernel,PASS,0.067 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP pool/avg_explicit_padding,PASS,-,0.00 MiB,0.00 MiB,1,0,0.008822 ms,78.027205 mW,688356.000000 pJ
pool/max_real_asymmetric_padding,PASS,0.063 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP pool/avg_include_pad,PASS,-,0.00 MiB,0.00 MiB,1,0,0.008506 ms,78.016929 mW,663612.000000 pJ
pool/max_same_upper,PASS,0.061 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP pool/avg_large_channels,PASS,-,0.04 MiB,0.02 MiB,1,0,0.235874 ms,78.004172 mW,18399156.000000 pJ
pool/max_stride2_multichannel,PASS,0.076 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP pool/avg_non_uniform_stride,PASS,-,0.00 MiB,0.00 MiB,1,0,0.014513 ms,78.016537 mW,1132254.000000 pJ
reduce_mean/4d_spatial,PASS,0.058 s,0.00 MiB,0.00 MiB,3,0,SKIP,SKIP,SKIP pool/avg_real_asymmetric_padding,PASS,-,0.00 MiB,0.00 MiB,1,0,0.025206 ms,78.024756 mW,1966692.000000 pJ
reduce_mean/4d_spatial_keepdims_0,PASS,0.068 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP pool/max_after_conv,PASS,-,0.00 MiB,0.00 MiB,5,4,0.012215 ms,99.115019 mW,1210689.960000 pJ
reduce_mean/after_conv,PASS,0.067 s,0.00 MiB,0.00 MiB,5,3,SKIP,SKIP,SKIP pool/max_basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.004160 ms,78.063462 mW,324744.000000 pJ
reduce_mean/all_axes_keepdims_0,PASS,0.057 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP pool/max_ceil_mode,PASS,-,0.00 MiB,0.00 MiB,1,0,0.001940 ms,78.074227 mW,151464.000000 pJ
reduce_mean/all_axes_keepdims_1,PASS,0.053 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP pool/max_global_style_kernel_equals_input,PASS,-,0.00 MiB,0.00 MiB,1,0,0.008443 ms,78.008528 mW,658626.000000 pJ
reduce_mean/basic,PASS,0.058 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP pool/max_non_square_kernel,PASS,-,0.00 MiB,0.00 MiB,1,0,0.013626 ms,78.017613 mW,1063068.000000 pJ
reduce_mean/channel_axis_nchw,PASS,0.063 s,0.03 MiB,0.02 MiB,4,0,SKIP,SKIP,SKIP pool/max_real_asymmetric_padding,PASS,-,0.00 MiB,0.00 MiB,1,0,0.010444 ms,78.034470 mW,814992.000000 pJ
reduce_mean/keepdims_0,PASS,0.064 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP pool/max_same_upper,PASS,-,0.00 MiB,0.00 MiB,1,0,0.008010 ms,78.035955 mW,625068.000000 pJ
reduce_mean/large_dimension_1024,PASS,0.066 s,0.01 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP pool/max_stride2_multichannel,PASS,-,0.00 MiB,0.00 MiB,1,0,0.015987 ms,78.018015 mW,1247274.000000 pJ
reduce_mean/legacy_axes_1_2_keepdims_1,PASS,0.069 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP reduce_mean/4d_spatial,PASS,-,0.00 MiB,0.00 MiB,3,0,0.000321 ms,92.448598 mW,29676.000000 pJ
reduce_mean/legacy_axis1_keepdims_0,PASS,0.067 s,0.00 MiB,0.00 MiB,9,0,SKIP,SKIP,SKIP reduce_mean/4d_spatial_keepdims_0,PASS,-,0.00 MiB,0.00 MiB,4,0,0.000655 ms,94.352672 mW,61801.000000 pJ
reduce_mean/legacy_axis1_keepdims_1,PASS,0.057 s,0.00 MiB,0.00 MiB,8,0,SKIP,SKIP,SKIP reduce_mean/after_conv,PASS,-,0.00 MiB,0.00 MiB,5,3,0.005342 ms,106.951089 mW,571332.720000 pJ
reduce_mean/legacy_empty_axes_noop,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reduce_mean/all_axes_keepdims_0,PASS,-,0.00 MiB,0.00 MiB,2,0,0.000391 ms,79.237852 mW,30982.000000 pJ
reduce_mean/legacy_nchw_spatial,PASS,0.059 s,0.00 MiB,0.00 MiB,3,0,SKIP,SKIP,SKIP reduce_mean/all_axes_keepdims_1,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000221 ms,78.217195 mW,17286.000000 pJ
reduce_mean/legacy_negative_axis,PASS,0.052 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP reduce_mean/basic,PASS,-,0.00 MiB,0.00 MiB,4,0,0.000373 ms,93.514745 mW,34881.000000 pJ
reduce_mean/legacy_reduce_all_keepdims_1,PASS,0.057 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reduce_mean/channel_axis_nchw,PASS,-,0.03 MiB,0.02 MiB,4,0,0.164926 ms,93.596631 mW,15436518.000000 pJ
reduce_mean/negative_axis,PASS,0.055 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP reduce_mean/keepdims_0,PASS,-,0.00 MiB,0.00 MiB,5,0,0.000748 ms,91.401070 mW,68368.000000 pJ
relu/4d,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reduce_mean/large_dimension_1024,PASS,-,0.01 MiB,0.00 MiB,1,0,0.002785 ms,78.017235 mW,217278.000000 pJ
relu/after_conv,PASS,0.062 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP reduce_mean/legacy_axes_1_2_keepdims_1,PASS,-,0.00 MiB,0.00 MiB,2,0,0.000271 ms,79.354244 mW,21505.000000 pJ
relu/after_gemm,PASS,0.062 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP reduce_mean/legacy_axis1_keepdims_0,PASS,-,0.00 MiB,0.00 MiB,9,0,0.001986 ms,92.501511 mW,183708.000000 pJ
relu/basic,PASS,0.062 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reduce_mean/legacy_axis1_keepdims_1,PASS,-,0.00 MiB,0.00 MiB,8,0,0.001373 ms,94.559359 mW,129830.000000 pJ
reshape/4d_to_2d_flatten,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reduce_mean/legacy_empty_axes_noop,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000221 ms,78.217195 mW,17286.000000 pJ
reshape/infer_dim_minus_one,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reduce_mean/legacy_nchw_spatial,PASS,-,0.00 MiB,0.00 MiB,3,0,0.000321 ms,92.448598 mW,29676.000000 pJ
reshape/same_rank,PASS,0.056 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reduce_mean/legacy_negative_axis,PASS,-,0.00 MiB,0.00 MiB,6,0,0.000553 ms,93.520796 mW,51717.000000 pJ
reshape/zero_copies_input_dim,PASS,0.077 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reduce_mean/legacy_reduce_all_keepdims_1,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000221 ms,78.217195 mW,17286.000000 pJ
resize/height_only,PASS,0.059 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP reduce_mean/negative_axis,PASS,-,0.00 MiB,0.00 MiB,6,0,0.000553 ms,93.520796 mW,51717.000000 pJ
resize/nearest_2x,PASS,0.066 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP relu/4d,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000521 ms,78.184261 mW,40734.000000 pJ
resize/nearest_downsample,PASS,0.062 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP relu/after_conv,PASS,-,0.00 MiB,0.00 MiB,4,3,0.005352 ms,107.891951 mW,577437.720000 pJ
resize/non_uniform,PASS,0.069 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP relu/after_gemm,PASS,-,0.01 MiB,0.01 MiB,5,4,0.007513 ms,105.158653 mW,790056.960000 pJ
resize/width_only,PASS,0.055 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP relu/basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000221 ms,78.217195 mW,17286.000000 pJ
resize/with_sizes,PASS,0.060 s,0.00 MiB,0.00 MiB,3,0,SKIP,SKIP,SKIP reshape/4d_to_2d_flatten,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000258 ms,78.279070 mW,20196.000000 pJ
sigmoid/4d,PASS,0.060 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reshape/infer_dim_minus_one,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000162 ms,78.296296 mW,12684.000000 pJ
sigmoid/after_gemm,PASS,0.059 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP reshape/same_rank,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000162 ms,78.296296 mW,12684.000000 pJ
sigmoid/basic,PASS,0.053 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP reshape/zero_copies_input_dim,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000162 ms,78.296296 mW,12684.000000 pJ
slice/2d_basic,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP resize/height_only,PASS,-,0.00 MiB,0.00 MiB,4,0,0.000693 ms,93.554113 mW,64833.000000 pJ
slice/after_conv,PASS,0.070 s,0.00 MiB,0.01 MiB,7,6,SKIP,SKIP,SKIP resize/nearest_2x,PASS,-,0.00 MiB,0.00 MiB,4,0,0.001173 ms,93.572890 mW,109761.000000 pJ
slice/default_axes,PASS,0.066 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP resize/nearest_downsample,PASS,-,0.00 MiB,0.00 MiB,2,0,0.000427 ms,79.449649 mW,33925.000000 pJ
slice/large_channel_1024,PASS,0.064 s,0.01 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP resize/non_uniform,PASS,-,0.00 MiB,0.00 MiB,6,0,0.001753 ms,93.575014 mW,164037.000000 pJ
slice/nchw_spatial_crop,PASS,0.055 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP resize/width_only,PASS,-,0.00 MiB,0.00 MiB,2,0,0.000667 ms,79.503748 mW,53029.000000 pJ
slice/negative_axis,PASS,0.060 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP resize/with_sizes,PASS,-,0.00 MiB,0.00 MiB,3,0,0.000797 ms,92.542033 mW,73756.000000 pJ
slice/negative_indices,PASS,0.053 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP sigmoid/4d,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000521 ms,78.184261 mW,40734.000000 pJ
slice/step2,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP sigmoid/after_gemm,PASS,-,0.01 MiB,0.01 MiB,5,4,0.007513 ms,105.158653 mW,790056.960000 pJ
softmax/3d_last_axis,PASS,0.050 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP sigmoid/basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000221 ms,78.217195 mW,17286.000000 pJ
softmax/basic,PASS,0.061 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP slice/2d_basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000242 ms,78.297521 mW,18948.000000 pJ
softmax/channel_axis,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP slice/after_conv,PASS,-,0.00 MiB,0.01 MiB,7,6,0.011296 ms,118.190765 mW,1335082.880000 pJ
softmax/large_dimension_1024,PASS,0.061 s,0.01 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP slice/default_axes,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000242 ms,78.297521 mW,18948.000000 pJ
softmax/negative_axis,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP slice/large_channel_1024,PASS,-,0.01 MiB,0.00 MiB,1,0,0.002832 ms,78.144068 mW,221304.000000 pJ
split/basic,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP slice/nchw_spatial_crop,PASS,-,0.00 MiB,0.00 MiB,1,0,0.001302 ms,78.239631 mW,101868.000000 pJ
split/equal_three_way,PASS,0.063 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP slice/negative_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000562 ms,78.298932 mW,44004.000000 pJ
split/negative_axis,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP slice/negative_indices,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000322 ms,78.298137 mW,25212.000000 pJ
split/uneven_channel_axis_4d,PASS,0.061 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP slice/step2,PASS,-,0.00 MiB,0.00 MiB,1,0,0.002042 ms,78.293830 mW,159876.000000 pJ
sub/after_gemm,PASS,0.064 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP softmax/3d_last_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,UNSUPPORTED,UNSUPPORTED,UNSUPPORTED
sub/basic,PASS,0.054 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP softmax/basic,PASS,-,0.00 MiB,0.00 MiB,1,0,UNSUPPORTED,UNSUPPORTED,UNSUPPORTED
sub/broadcast_row,PASS,0.064 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP softmax/channel_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,UNSUPPORTED,UNSUPPORTED,UNSUPPORTED
sub/channel_broadcast_1024,PASS,0.063 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP softmax/large_dimension_1024,PASS,-,0.01 MiB,0.01 MiB,1,0,UNSUPPORTED,UNSUPPORTED,UNSUPPORTED
sub/constant_lhs_broadcast,PASS,0.054 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP softmax/negative_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,UNSUPPORTED,UNSUPPORTED,UNSUPPORTED
sub/leading_dimension_broadcast,PASS,0.054 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP split/basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000403 ms,78.297767 mW,31554.000000 pJ
split/equal_three_way,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000564 ms,78.297872 mW,44160.000000 pJ
split/negative_axis,PASS,-,0.00 MiB,0.00 MiB,1,0,0.001083 ms,78.288089 mW,84786.000000 pJ
split/uneven_channel_axis_4d,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000242 ms,78.297521 mW,18948.000000 pJ
sub/after_gemm,PASS,-,0.01 MiB,0.01 MiB,5,4,0.007784 ms,104.703618 mW,815012.960000 pJ
sub/basic,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
sub/broadcast_row,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
sub/channel_broadcast_1024,PASS,-,0.02 MiB,0.01 MiB,1,0,0.006913 ms,78.118038 mW,540030.000000 pJ
sub/constant_lhs_broadcast,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000322 ms,78.223602 mW,25188.000000 pJ
sub/leading_dimension_broadcast,PASS,-,0.00 MiB,0.00 MiB,1,0,0.000323 ms,78.222910 mW,25266.000000 pJ
1 Operation Result Compile Host mem Cores mem Cores Xbars Latency Power Energy
2 add/after_gemm PASS 0.063 s - 0.01 MiB 0.01 MiB 5 4 SKIP 0.007784 ms SKIP 104.703618 mW SKIP 815012.960000 pJ
3 add/basic PASS 0.057 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
4 add/broadcast_row PASS 0.057 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
5 add/channel_broadcast_1024 PASS 0.061 s - 0.02 MiB 0.01 MiB 1 0 SKIP 0.006913 ms SKIP 78.118038 mW SKIP 540030.000000 pJ
6 add/leading_dimension_broadcast PASS 0.057 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
7 concat/channel_axis PASS 0.069 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000457 ms SKIP 78.157549 mW SKIP 35718.000000 pJ
8 concat/negative_axis PASS 0.061 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.001043 ms SKIP 78.092042 mW SKIP 81450.000000 pJ
9 concat/three_inputs_channel_axis PASS 0.058 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000644 ms SKIP 78.149068 mW SKIP 50328.000000 pJ
10 conv/batch_2 PASS 0.066 s - 0.00 MiB 0.00 MiB 2 2 SKIP 0.013694 ms SKIP 82.623885 mW SKIP 1131451.480000 pJ
11 conv/batch_4_pointwise PASS 0.065 s - 0.00 MiB 0.01 MiB 5 4 SKIP 0.003932 ms SKIP 116.078576 mW SKIP 456420.960000 pJ
12 conv/depthwise_1024_channels conv/conv_1x3x224x224_w64x3x7x7_b64 PASS 0.083 s - 0.19 MiB 24.14 MiB 0.38 MiB 61.87 MiB 129 168 128 169 SKIP 38.414551 ms SKIP 185.256473 mW SKIP 7116544212.120002 pJ
13 conv/depthwise_grouped conv/depthwise_1024_channels PASS 0.071 s - 0.01 MiB 0.19 MiB 0.00 MiB 0.38 MiB 5 129 4 128 SKIP 0.220751 ms SKIP 178.454307 mW SKIP 39393966.720000 pJ
14 conv/dilated_3x3 conv/depthwise_grouped PASS 0.068 s - 0.01 MiB 0.01 MiB 0.00 MiB 10 5 9 4 SKIP 0.006234 ms SKIP 107.776541 mW SKIP 671878.960000 pJ
15 conv/dynamic conv/dilated_3x3 PASS 0.065 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.01 MiB 5 10 0 9 SKIP 0.008713 ms SKIP 118.767263 mW SKIP 1034819.160000 pJ
16 conv/explicit_padding conv/dynamic PASS 0.062 s - 0.01 MiB 0.00 MiB 0.02 MiB 0.00 MiB 17 5 16 0 SKIP 0.001835 ms SKIP 92.281199 mW SKIP 169336.000000 pJ
17 conv/grouped_many_groups conv/explicit_padding PASS 0.442 s - 0.05 MiB 0.01 MiB 0.09 MiB 0.02 MiB 65 17 64 16 SKIP 0.010007 ms SKIP 145.338047 mW SKIP 1454397.840000 pJ
18 conv/grouped_two_groups conv/grouped_many_groups PASS 0.060 s - 0.00 MiB 0.05 MiB 0.00 MiB 0.09 MiB 3 65 2 64 SKIP 0.181897 ms SKIP 142.207471 mW SKIP 25867112.360000 pJ
19 conv/huge_pointwise_1024 conv/grouped_two_groups PASS 0.156 s - 0.01 MiB 0.00 MiB 0.11 MiB 0.00 MiB 73 3 64 2 SKIP 0.005361 ms SKIP 101.457653 mW SKIP 543914.480000 pJ
20 conv/huge_pointwise_1024_dynamic conv/huge_pointwise_1024 PASS 0.079 s - 8.04 MiB 0.01 MiB 12.61 MiB 0.11 MiB 168 73 0 64 SKIP 0.015615 ms SKIP 249.545140 mW SKIP 3896647.360000 pJ
21 conv/kernel_3x3 conv/huge_pointwise_1024_dynamic PASS 0.062 s - 0.01 MiB 8.04 MiB 0.01 MiB 12.61 MiB 10 168 9 0 SKIP 2.627964 ms SKIP 169.518697 mW SKIP 445489032.000000 pJ
22 conv/kernel_equals_input_spatial conv/kernel_3x3 PASS 0.064 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.01 MiB 2 10 2 9 SKIP 0.007186 ms SKIP 123.801859 mW SKIP 889640.160000 pJ
23 conv/large_input_channels_1x1 conv/kernel_equals_input_spatial PASS 0.089 s - 0.01 MiB 0.00 MiB 0.02 MiB 0.00 MiB 9 2 8 2 SKIP 0.004639 ms SKIP 89.607562 mW SKIP 415689.480000 pJ
24 conv/large_output_channels_1x1 conv/large_input_channels_1x1 PASS 0.098 s - 0.01 MiB 0.02 MiB 17 9 8 SKIP 0.007648 ms SKIP 117.824519 mW SKIP 901121.920000 pJ
25 conv/large_spatial conv/large_output_channels_1x1 PASS 0.069 s - 0.01 MiB 0.04 MiB 0.02 MiB 37 17 36 8 SKIP 0.008871 ms SKIP 128.442782 mW SKIP 1139415.920000 pJ
26 conv/multi_channel conv/large_spatial PASS 0.066 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.04 MiB 4 37 3 36 SKIP 0.017018 ms SKIP 172.073372 mW SKIP 2928344.640000 pJ
27 conv/non_square_kernel_1x3 conv/multi_channel PASS 0.061 s - 0.00 MiB 0.00 MiB 3 4 2 3 SKIP 0.006482 ms SKIP 105.683542 mW SKIP 685040.720000 pJ
28 conv/non_square_kernel_3x1 conv/non_square_kernel_1x3 PASS 0.064 s - 0.00 MiB 0.00 MiB 3 2 SKIP 0.006842 ms SKIP 99.349968 mW SKIP 679752.480000 pJ
29 conv/non_uniform_stride conv/non_square_kernel_3x1 PASS 0.062 s - 0.00 MiB 0.00 MiB 4 3 3 2 SKIP 0.013484 ms SKIP 95.889683 mW SKIP 1292976.480000 pJ
30 conv/pointwise_1x1 conv/non_uniform_stride PASS 0.059 s - 0.00 MiB 0.00 MiB 1 4 1 3 SKIP 0.007601 ms SKIP 104.048772 mW SKIP 790874.720000 pJ
31 conv/pointwise_tiled_chain conv/output_channel_grouping_minimal PASS 0.604 s - 0.01 MiB 0.10 MiB 0.04 MiB 0.34 MiB 20 131 80 128 SKIP 0.258453 ms SKIP 170.730913 mW SKIP 44125916.720000 pJ
32 conv/real_asymmetric_padding conv/pointwise_1x1 PASS 0.060 s - 0.01 MiB 0.00 MiB 0.03 MiB 0.00 MiB 29 1 28 1 SKIP 0.012303 ms SKIP 80.244188 mW SKIP 987244.240000 pJ
33 conv/relu_conv_store conv/pointwise_tiled_chain PASS 0.091 s - 0.16 MiB 0.01 MiB 0.67 MiB 0.04 MiB 168 20 184 80 SKIP 0.041886 ms SKIP 153.880896 mW SKIP 6445455.200000 pJ
34 conv/same_lower_3x3 conv/real_asymmetric_padding PASS 0.078 s - 0.01 MiB 0.02 MiB 0.03 MiB 26 29 25 28 SKIP 0.014457 ms SKIP 153.669968 mW SKIP 2221606.720000 pJ
35 conv/same_padding_3x3 conv/relu_conv_store PASS 0.070 s - 0.01 MiB 0.16 MiB 0.02 MiB 0.67 MiB 26 168 25 184 SKIP 0.562898 ms SKIP 183.084489 mW SKIP 103057892.800000 pJ
36 conv/simple conv/same_lower_3x3 PASS 0.064 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.02 MiB 1 26 1 25 SKIP 0.013331 ms SKIP 166.154752 mW SKIP 2215009.000000 pJ
37 conv/stride_2 conv/same_padding_3x3 PASS 0.064 s - 0.01 MiB 0.00 MiB 0.02 MiB 5 26 4 25 SKIP 0.013331 ms SKIP 166.154752 mW SKIP 2215009.000000 pJ
38 conv/with_bias_3x3 conv/simple PASS 0.067 s - 0.00 MiB 0.01 MiB 0.00 MiB 4 1 3 1 SKIP 0.004301 ms SKIP 83.833583 mW SKIP 360568.240000 pJ
39 conv/with_constant conv/strategy_depthwise_16 PASS 0.070 s - 0.00 MiB 0.06 MiB 0.00 MiB 0.35 MiB 1 168 1 168 SKIP 0.335115 ms SKIP 197.936467 mW SKIP 66331479.080000 pJ
40 conv/without_kernel_shape_attr conv/strategy_input_k_tiled PASS 0.069 s - 0.01 MiB 0.08 MiB 0.01 MiB 0.27 MiB 10 109 9 108 SKIP 0.353736 ms SKIP 170.812713 mW SKIP 60422605.920000 pJ
41 conv/yolo11n_depthwise_head conv/strategy_output_channel_tiled PASS 1.482 s - 8.66 MiB 0.03 MiB 34.24 MiB 0.16 MiB 168 74 255 72 SKIP 0.091463 ms SKIP 155.743189 mW SKIP 14244739.280000 pJ
42 conv/yolo11n_heavy conv/strategy_streamed_packed PASS 0.431 s - 4.82 MiB 3.34 MiB 19.10 MiB 7.89 MiB 161 168 800 168 SKIP 9.353833 ms SKIP 179.858301 mW SKIP 1682364509.560000 pJ
43 conv/yolo11n_stem conv/strategy_streamed_patch PASS 0.783 s - 12.86 MiB 0.34 MiB 37.59 MiB 1.32 MiB 168 488 168 SKIP 1.904655 ms SKIP 181.910449 mW SKIP 346476645.640000 pJ
44 div/after_gemm conv/strategy_tiled_2d PASS 0.072 s - 0.01 MiB 0.11 MiB 0.01 MiB 0.44 MiB 5 168 4 168 SKIP 0.415594 ms SKIP 182.127047 mW SKIP 75690907.840000 pJ
45 div/basic conv/stride_2 PASS 0.061 s - 0.00 MiB 0.01 MiB 0.00 MiB 1 5 0 4 SKIP 0.005237 ms SKIP 110.780019 mW SKIP 580154.960000 pJ
46 div/channel_broadcast_1024 conv/with_bias_3x3 PASS 0.066 s - 0.02 MiB 0.00 MiB 0.01 MiB 1 4 0 3 SKIP 0.007452 ms SKIP 104.162738 mW SKIP 776220.720000 pJ
47 div/leading_dimension_broadcast conv/with_constant PASS 0.059 s - 0.00 MiB 0.00 MiB 1 0 1 SKIP 0.006622 ms SKIP 81.738182 mW SKIP 541270.240000 pJ
48 div/runtime_scalar_rhs conv/without_kernel_shape_attr PASS 0.056 s - 0.02 MiB 0.01 MiB 0.01 MiB 1 10 0 9 SKIP 0.007186 ms SKIP 123.801859 mW SKIP 889640.160000 pJ
49 div/scalar_constant conv/yolo11n_depthwise_head PASS 0.073 s - 0.00 MiB 8.66 MiB 0.00 MiB 34.24 MiB 1 168 0 255 SKIP 42.701404 ms SKIP 200.519161 mW SKIP 8562449708.000010 pJ
50 gather/3d_input_axis1 conv/yolo11n_heavy PASS 0.060 s - 0.00 MiB 4.82 MiB 0.00 MiB 19.10 MiB 1 161 0 800 SKIP 8.535404 ms SKIP 350.863768 mW SKIP 2994764012.000010 pJ
51 gather/axis0_matrix_indices conv/yolo11n_stem PASS 0.056 s - 0.00 MiB 12.86 MiB 0.00 MiB 37.59 MiB 1 168 0 488 SKIP 14.239985 ms SKIP 301.233376 mW SKIP 4289558753.000010 pJ
52 gather/axis1 div/after_gemm PASS 0.058 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.01 MiB 1 5 0 4 SKIP 0.007784 ms SKIP 104.703618 mW SKIP 815012.960000 pJ
53 gather/negative_axis div/basic PASS 0.058 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
54 gather/negative_indices div/channel_broadcast_1024 PASS 0.063 s - 0.00 MiB 0.02 MiB 0.00 MiB 0.01 MiB 1 0 SKIP 0.006913 ms SKIP 78.118038 mW SKIP 540030.000000 pJ
55 gemm/alpha_beta div/leading_dimension_broadcast PASS 0.067 s - 0.01 MiB 0.00 MiB 0.01 MiB 0.00 MiB 5 1 4 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
56 gemm/bias_rank2_broadcast div/runtime_scalar_rhs PASS 0.060 s - 0.00 MiB 0.02 MiB 0.01 MiB 5 1 4 0 SKIP 0.006913 ms SKIP 78.118038 mW SKIP 540030.000000 pJ
57 gemm/dynamic div/scalar_constant PASS 0.064 s - 0.00 MiB 0.00 MiB 5 1 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
58 gemm/dynamic_alpha gather/3d_input_axis1 PASS 0.062 s - 0.00 MiB 0.00 MiB 5 1 0 SKIP 0.000589 ms SKIP 78.081494 mW SKIP 45990.000000 pJ
59 gemm/dynamic_beta gather/axis0_matrix_indices PASS 0.060 s - 0.00 MiB 0.00 MiB 5 1 0 SKIP 0.000697 ms SKIP 78.068867 mW SKIP 54414.000000 pJ
60 gemm/dynamic_bias gather/axis1 PASS 0.058 s - 0.00 MiB 0.00 MiB 5 1 0 SKIP 0.000801 ms SKIP 78.059925 mW SKIP 62526.000000 pJ
61 gemm/dynamic_bias_alpha_beta gather/negative_axis PASS 0.067 s - 0.00 MiB 0.00 MiB 5 1 0 SKIP 0.001437 ms SKIP 78.033403 mW SKIP 112134.000000 pJ
62 gemm/dynamic_transB gather/negative_indices PASS 0.062 s - 0.00 MiB 0.00 MiB 5 1 0 SKIP 0.000376 ms SKIP 78.127660 mW SKIP 29376.000000 pJ
63 gemm/huge_1024 gemm/alpha_beta PASS 0.147 s - 0.01 MiB 0.10 MiB 0.01 MiB 73 5 64 4 SKIP 0.007456 ms SKIP 105.272125 mW SKIP 784908.960000 pJ
64 gemm/large gemm/bias_rank2_broadcast PASS 0.068 s - 0.02 MiB 0.00 MiB 0.03 MiB 0.01 MiB 17 5 16 4 SKIP 0.007072 ms SKIP 105.979208 mW SKIP 749484.960000 pJ
65 gemm/large_k_small_n gemm/dynamic PASS 0.095 s - 0.01 MiB 0.00 MiB 0.01 MiB 0.00 MiB 9 5 8 0 SKIP 0.002421 ms SKIP 91.480793 mW SKIP 221475.000000 pJ
66 gemm/non_square gemm/dynamic_alpha PASS 0.063 s - 0.00 MiB 0.01 MiB 0.00 MiB 5 4 0 SKIP 0.003262 ms SKIP 91.415696 mW SKIP 298198.000000 pJ
67 gemm/scalar_bias gemm/dynamic_beta PASS 0.060 s - 0.00 MiB 0.01 MiB 0.00 MiB 5 4 0 SKIP 0.004365 ms SKIP 91.316151 mW SKIP 398595.000000 pJ
68 gemm/simple gemm/dynamic_bias PASS 0.072 s - 0.03 MiB 0.00 MiB 0.08 MiB 0.00 MiB 42 5 40 0 SKIP 0.002665 ms SKIP 91.445779 mW SKIP 243703.000000 pJ
69 gemm/small gemm/dynamic_bias_alpha_beta PASS 0.065 s - 0.00 MiB 0.00 MiB 2 5 2 0 SKIP 0.005629 ms SKIP 91.279268 mW SKIP 513811.000000 pJ
70 gemm/small_k_large_n gemm/dynamic_transB PASS 0.097 s - 0.01 MiB 0.00 MiB 0.02 MiB 0.00 MiB 17 5 8 0 SKIP 0.001301 ms SKIP 91.378171 mW SKIP 118883.000000 pJ
71 gemm/transA gemm/huge_1024 PASS 0.064 s - 0.00 MiB 0.01 MiB 0.01 MiB 0.10 MiB 5 73 4 64 SKIP 0.017522 ms SKIP 215.037402 mW SKIP 3767885.360000 pJ
72 gemm/transA_transB gemm/large PASS 0.069 s - 0.00 MiB 0.02 MiB 0.01 MiB 0.03 MiB 5 17 4 16 SKIP 0.011229 ms SKIP 140.152181 mW SKIP 1573768.840000 pJ
73 gemm/transB gemm/large_k_small_n PASS 0.062 s - 0.00 MiB 0.01 MiB 0.01 MiB 5 9 4 8 SKIP 0.004748 ms SKIP 133.481449 mW SKIP 633769.920000 pJ
74 gemm/transB_with_bias gemm/non_square PASS 0.055 s - 0.01 MiB 0.00 MiB 0.01 MiB 5 4 SKIP 0.003527 ms SKIP 118.958310 mW SKIP 419565.960000 pJ
75 gemm/with_bias gemm/scalar_bias PASS 0.067 s - 0.01 MiB 0.00 MiB 0.01 MiB 5 4 SKIP 0.007072 ms SKIP 105.979208 mW SKIP 749484.960000 pJ
76 gemv/constant gemm/simple PASS 0.064 s - 0.00 MiB 0.03 MiB 0.00 MiB 0.08 MiB 0 42 0 40 SKIP 0.021640 ms SKIP 151.774196 mW SKIP 3284393.600000 pJ
77 gemv/simple gemm/small PASS 0.069 s - 0.00 MiB 0.01 MiB 0.00 MiB 6 2 4 2 SKIP 0.004420 ms SKIP 90.144000 mW SKIP 398436.480000 pJ
78 gemv/with_heterogeneous_constant gemm/small_k_large_n PASS 0.066 s - 0.00 MiB 0.01 MiB 0.01 MiB 0.02 MiB 6 17 4 8 SKIP 0.007962 ms SKIP 131.005014 mW SKIP 1043061.920000 pJ
79 gemv/with_homogeneous_constant gemm/transA PASS 0.070 s - 0.00 MiB 0.01 MiB 6 5 4 SKIP 0.005762 ms SKIP 109.140743 mW SKIP 628868.960000 pJ
80 gemv/with_scalar_constant gemm/transA_transB PASS 0.070 s - 0.00 MiB 0.01 MiB 6 5 4 SKIP 0.005762 ms SKIP 109.140743 mW SKIP 628868.960000 pJ
81 matmul/basic gemm/transB PASS 0.062 s - 0.00 MiB 0.00 MiB 0.01 MiB 2 5 2 4 SKIP 0.003527 ms SKIP 118.958310 mW SKIP 419565.960000 pJ
82 matmul/batched_3d gemm/transB_with_bias PASS 0.066 s - 0.00 MiB 0.01 MiB 0.01 MiB 5 4 SKIP 0.005046 ms SKIP 110.546762 mW SKIP 557818.960000 pJ
83 matmul/batched_3d_dynamic gemm/with_bias PASS 0.057 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.01 MiB 4 5 0 4 SKIP 0.005562 ms SKIP 108.767882 mW SKIP 604966.960000 pJ
84 matmul/batched_left_constant gemv/constant PASS 0.070 s - 0.00 MiB 0.02 MiB 0.00 MiB 9 0 8 0 SKIP 0.000000 ms SKIP 2.000000 mW SKIP 0.000000 pJ
85 matmul/batched_lhs_broadcast gemv/simple PASS 0.063 s - 0.00 MiB 0.01 MiB 5 6 4 SKIP 0.005160 ms SKIP 111.150380 mW SKIP 573535.960000 pJ
86 matmul/batched_rhs_broadcast gemv/with_heterogeneous_constant PASS 0.062 s - 0.00 MiB 0.01 MiB 5 6 4 SKIP 0.005549 ms SKIP 109.816536 mW SKIP 609371.960000 pJ
87 matmul/dynamic gemv/with_homogeneous_constant PASS 0.060 s - 0.00 MiB 0.00 MiB 0.01 MiB 5 6 0 4 SKIP 0.005549 ms SKIP 109.816536 mW SKIP 609371.960000 pJ
88 matmul/huge_1024 gemv/with_scalar_constant PASS 0.145 s - 0.01 MiB 0.00 MiB 0.10 MiB 0.01 MiB 73 6 64 4 SKIP 0.005549 ms SKIP 109.816536 mW SKIP 609371.960000 pJ
89 matmul/left_constant matmul/basic PASS 0.051 s - 0.00 MiB 0.01 MiB 0.00 MiB 5 2 4 2 SKIP 0.004420 ms SKIP 90.144000 mW SKIP 398436.480000 pJ
90 matmul/matrix_vector matmul/batched_3d PASS 0.095 s - 0.52 MiB 0.00 MiB 0.78 MiB 0.01 MiB 168 5 173 4 SKIP 0.005958 ms SKIP 108.588949 mW SKIP 646972.960000 pJ
91 matmul/vector_matrix matmul/batched_3d_dynamic PASS 0.087 s - 0.01 MiB 0.00 MiB 0.01 MiB 0.00 MiB 9 4 8 0 SKIP 0.001822 ms SKIP 92.192645 mW SKIP 167975.000000 pJ
92 matmul/yolo_attention matmul/batched_left_constant PASS 0.385 s - 1.02 MiB 0.00 MiB 43.44 MiB 0.02 MiB 168 9 0 8 SKIP 0.008822 ms SKIP 114.385164 mW SKIP 1009105.920000 pJ
93 mul/after_conv matmul/batched_lhs_broadcast PASS 0.055 s - 0.00 MiB 0.00 MiB 0.01 MiB 4 5 3 4 SKIP 0.005681 ms SKIP 109.389361 mW SKIP 621440.960000 pJ
94 mul/basic matmul/batched_rhs_broadcast PASS 0.055 s - 0.00 MiB 0.00 MiB 0.01 MiB 1 5 0 4 SKIP 0.005958 ms SKIP 108.588949 mW SKIP 646972.960000 pJ
95 mul/channel_broadcast_1024 matmul/dynamic PASS 0.058 s - 0.02 MiB 0.00 MiB 0.01 MiB 0.00 MiB 1 5 0 SKIP 0.001621 ms SKIP 91.421962 mW SKIP 148195.000000 pJ
96 mul/leading_dimension_broadcast matmul/huge_1024 PASS 0.059 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.10 MiB 1 73 0 64 SKIP 0.017522 ms SKIP 215.037402 mW SKIP 3767885.360000 pJ
97 mul/scalar_constant matmul/left_constant PASS 0.056 s - 0.00 MiB 0.00 MiB 0.01 MiB 1 5 0 4 SKIP 0.005853 ms SKIP 108.861944 mW SKIP 637168.960000 pJ
98 pool/avg_basic matmul/matrix_vector PASS 0.063 s - 0.00 MiB 0.52 MiB 0.00 MiB 0.78 MiB 1 168 0 173 SKIP 0.384660 ms SKIP 202.131271 mW SKIP 77751814.880000 pJ
99 pool/avg_ceil_mode matmul/vector_matrix PASS 0.063 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.01 MiB 1 9 0 8 SKIP 0.007409 ms SKIP 118.680243 mW SKIP 879301.920000 pJ
100 pool/avg_explicit_padding matmul/yolo_attention PASS 0.061 s - 0.00 MiB 1.02 MiB 0.00 MiB 43.44 MiB 1 168 0 SKIP 8.151445 ms SKIP 170.003707 mW SKIP 1385775865.000000 pJ
101 pool/avg_include_pad mul/after_conv PASS 0.049 s - 0.00 MiB 0.00 MiB 1 4 0 3 SKIP 0.005453 ms SKIP 107.639046 mW SKIP 586955.720000 pJ
102 pool/avg_large_channels mul/after_conv_scalar_constant PASS 0.059 s - 0.04 MiB 0.00 MiB 0.02 MiB 0.00 MiB 1 4 0 3 SKIP 0.005453 ms SKIP 107.639046 mW SKIP 586955.720000 pJ
103 pool/avg_non_uniform_stride mul/basic PASS 0.066 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
104 pool/avg_real_asymmetric_padding mul/channel_broadcast_1024 PASS 0.055 s - 0.00 MiB 0.02 MiB 0.00 MiB 0.01 MiB 1 0 SKIP 0.006913 ms SKIP 78.118038 mW SKIP 540030.000000 pJ
105 pool/max_after_conv mul/leading_dimension_broadcast PASS 0.069 s - 0.00 MiB 0.00 MiB 5 1 4 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
106 pool/max_basic mul/scalar_constant PASS 0.064 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000323 ms SKIP 78.222910 mW SKIP 25266.000000 pJ
107 pool/max_ceil_mode pool/avg_basic PASS 0.058 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.011939 ms SKIP 78.022112 mW SKIP 931506.000000 pJ
108 pool/max_global_style_kernel_equals_input pool/avg_ceil_mode PASS 0.063 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.004359 ms SKIP 78.033035 mW SKIP 340146.000000 pJ
109 pool/max_non_square_kernel pool/avg_explicit_padding PASS 0.067 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.008822 ms SKIP 78.027205 mW SKIP 688356.000000 pJ
110 pool/max_real_asymmetric_padding pool/avg_include_pad PASS 0.063 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.008506 ms SKIP 78.016929 mW SKIP 663612.000000 pJ
111 pool/max_same_upper pool/avg_large_channels PASS 0.061 s - 0.00 MiB 0.04 MiB 0.00 MiB 0.02 MiB 1 0 SKIP 0.235874 ms SKIP 78.004172 mW SKIP 18399156.000000 pJ
112 pool/max_stride2_multichannel pool/avg_non_uniform_stride PASS 0.076 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.014513 ms SKIP 78.016537 mW SKIP 1132254.000000 pJ
113 reduce_mean/4d_spatial pool/avg_real_asymmetric_padding PASS 0.058 s - 0.00 MiB 0.00 MiB 3 1 0 SKIP 0.025206 ms SKIP 78.024756 mW SKIP 1966692.000000 pJ
114 reduce_mean/4d_spatial_keepdims_0 pool/max_after_conv PASS 0.068 s - 0.00 MiB 0.00 MiB 4 5 0 4 SKIP 0.012215 ms SKIP 99.115019 mW SKIP 1210689.960000 pJ
115 reduce_mean/after_conv pool/max_basic PASS 0.067 s - 0.00 MiB 0.00 MiB 5 1 3 0 SKIP 0.004160 ms SKIP 78.063462 mW SKIP 324744.000000 pJ
116 reduce_mean/all_axes_keepdims_0 pool/max_ceil_mode PASS 0.057 s - 0.00 MiB 0.00 MiB 2 1 0 SKIP 0.001940 ms SKIP 78.074227 mW SKIP 151464.000000 pJ
117 reduce_mean/all_axes_keepdims_1 pool/max_global_style_kernel_equals_input PASS 0.053 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.008443 ms SKIP 78.008528 mW SKIP 658626.000000 pJ
118 reduce_mean/basic pool/max_non_square_kernel PASS 0.058 s - 0.00 MiB 0.00 MiB 4 1 0 SKIP 0.013626 ms SKIP 78.017613 mW SKIP 1063068.000000 pJ
119 reduce_mean/channel_axis_nchw pool/max_real_asymmetric_padding PASS 0.063 s - 0.03 MiB 0.00 MiB 0.02 MiB 0.00 MiB 4 1 0 SKIP 0.010444 ms SKIP 78.034470 mW SKIP 814992.000000 pJ
120 reduce_mean/keepdims_0 pool/max_same_upper PASS 0.064 s - 0.00 MiB 0.00 MiB 5 1 0 SKIP 0.008010 ms SKIP 78.035955 mW SKIP 625068.000000 pJ
121 reduce_mean/large_dimension_1024 pool/max_stride2_multichannel PASS 0.066 s - 0.01 MiB 0.00 MiB 0.00 MiB 1 0 SKIP 0.015987 ms SKIP 78.018015 mW SKIP 1247274.000000 pJ
122 reduce_mean/legacy_axes_1_2_keepdims_1 reduce_mean/4d_spatial PASS 0.069 s - 0.00 MiB 0.00 MiB 2 3 0 SKIP 0.000321 ms SKIP 92.448598 mW SKIP 29676.000000 pJ
123 reduce_mean/legacy_axis1_keepdims_0 reduce_mean/4d_spatial_keepdims_0 PASS 0.067 s - 0.00 MiB 0.00 MiB 9 4 0 SKIP 0.000655 ms SKIP 94.352672 mW SKIP 61801.000000 pJ
124 reduce_mean/legacy_axis1_keepdims_1 reduce_mean/after_conv PASS 0.057 s - 0.00 MiB 0.00 MiB 8 5 0 3 SKIP 0.005342 ms SKIP 106.951089 mW SKIP 571332.720000 pJ
125 reduce_mean/legacy_empty_axes_noop reduce_mean/all_axes_keepdims_0 PASS 0.058 s - 0.00 MiB 0.00 MiB 1 2 0 SKIP 0.000391 ms SKIP 79.237852 mW SKIP 30982.000000 pJ
126 reduce_mean/legacy_nchw_spatial reduce_mean/all_axes_keepdims_1 PASS 0.059 s - 0.00 MiB 0.00 MiB 3 1 0 SKIP 0.000221 ms SKIP 78.217195 mW SKIP 17286.000000 pJ
127 reduce_mean/legacy_negative_axis reduce_mean/basic PASS 0.052 s - 0.00 MiB 0.00 MiB 6 4 0 SKIP 0.000373 ms SKIP 93.514745 mW SKIP 34881.000000 pJ
128 reduce_mean/legacy_reduce_all_keepdims_1 reduce_mean/channel_axis_nchw PASS 0.057 s - 0.00 MiB 0.03 MiB 0.00 MiB 0.02 MiB 1 4 0 SKIP 0.164926 ms SKIP 93.596631 mW SKIP 15436518.000000 pJ
129 reduce_mean/negative_axis reduce_mean/keepdims_0 PASS 0.055 s - 0.00 MiB 0.00 MiB 6 5 0 SKIP 0.000748 ms SKIP 91.401070 mW SKIP 68368.000000 pJ
130 relu/4d reduce_mean/large_dimension_1024 PASS 0.059 s - 0.00 MiB 0.01 MiB 0.00 MiB 1 0 SKIP 0.002785 ms SKIP 78.017235 mW SKIP 217278.000000 pJ
131 relu/after_conv reduce_mean/legacy_axes_1_2_keepdims_1 PASS 0.062 s - 0.00 MiB 0.00 MiB 4 2 3 0 SKIP 0.000271 ms SKIP 79.354244 mW SKIP 21505.000000 pJ
132 relu/after_gemm reduce_mean/legacy_axis1_keepdims_0 PASS 0.062 s - 0.01 MiB 0.00 MiB 0.01 MiB 0.00 MiB 5 9 4 0 SKIP 0.001986 ms SKIP 92.501511 mW SKIP 183708.000000 pJ
133 relu/basic reduce_mean/legacy_axis1_keepdims_1 PASS 0.062 s - 0.00 MiB 0.00 MiB 1 8 0 SKIP 0.001373 ms SKIP 94.559359 mW SKIP 129830.000000 pJ
134 reshape/4d_to_2d_flatten reduce_mean/legacy_empty_axes_noop PASS 0.059 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000221 ms SKIP 78.217195 mW SKIP 17286.000000 pJ
135 reshape/infer_dim_minus_one reduce_mean/legacy_nchw_spatial PASS 0.059 s - 0.00 MiB 0.00 MiB 1 3 0 SKIP 0.000321 ms SKIP 92.448598 mW SKIP 29676.000000 pJ
136 reshape/same_rank reduce_mean/legacy_negative_axis PASS 0.056 s - 0.00 MiB 0.00 MiB 1 6 0 SKIP 0.000553 ms SKIP 93.520796 mW SKIP 51717.000000 pJ
137 reshape/zero_copies_input_dim reduce_mean/legacy_reduce_all_keepdims_1 PASS 0.077 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000221 ms SKIP 78.217195 mW SKIP 17286.000000 pJ
138 resize/height_only reduce_mean/negative_axis PASS 0.059 s - 0.00 MiB 0.00 MiB 4 6 0 SKIP 0.000553 ms SKIP 93.520796 mW SKIP 51717.000000 pJ
139 resize/nearest_2x relu/4d PASS 0.066 s - 0.00 MiB 0.00 MiB 4 1 0 SKIP 0.000521 ms SKIP 78.184261 mW SKIP 40734.000000 pJ
140 resize/nearest_downsample relu/after_conv PASS 0.062 s - 0.00 MiB 0.00 MiB 2 4 0 3 SKIP 0.005352 ms SKIP 107.891951 mW SKIP 577437.720000 pJ
141 resize/non_uniform relu/after_gemm PASS 0.069 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.01 MiB 6 5 0 4 SKIP 0.007513 ms SKIP 105.158653 mW SKIP 790056.960000 pJ
142 resize/width_only relu/basic PASS 0.055 s - 0.00 MiB 0.00 MiB 2 1 0 SKIP 0.000221 ms SKIP 78.217195 mW SKIP 17286.000000 pJ
143 resize/with_sizes reshape/4d_to_2d_flatten PASS 0.060 s - 0.00 MiB 0.00 MiB 3 1 0 SKIP 0.000258 ms SKIP 78.279070 mW SKIP 20196.000000 pJ
144 sigmoid/4d reshape/infer_dim_minus_one PASS 0.060 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000162 ms SKIP 78.296296 mW SKIP 12684.000000 pJ
145 sigmoid/after_gemm reshape/same_rank PASS 0.059 s - 0.01 MiB 0.00 MiB 0.01 MiB 0.00 MiB 5 1 4 0 SKIP 0.000162 ms SKIP 78.296296 mW SKIP 12684.000000 pJ
146 sigmoid/basic reshape/zero_copies_input_dim PASS 0.053 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000162 ms SKIP 78.296296 mW SKIP 12684.000000 pJ
147 slice/2d_basic resize/height_only PASS 0.058 s - 0.00 MiB 0.00 MiB 1 4 0 SKIP 0.000693 ms SKIP 93.554113 mW SKIP 64833.000000 pJ
148 slice/after_conv resize/nearest_2x PASS 0.070 s - 0.00 MiB 0.01 MiB 0.00 MiB 7 4 6 0 SKIP 0.001173 ms SKIP 93.572890 mW SKIP 109761.000000 pJ
149 slice/default_axes resize/nearest_downsample PASS 0.066 s - 0.00 MiB 0.00 MiB 1 2 0 SKIP 0.000427 ms SKIP 79.449649 mW SKIP 33925.000000 pJ
150 slice/large_channel_1024 resize/non_uniform PASS 0.064 s - 0.01 MiB 0.00 MiB 0.00 MiB 1 6 0 SKIP 0.001753 ms SKIP 93.575014 mW SKIP 164037.000000 pJ
151 slice/nchw_spatial_crop resize/width_only PASS 0.055 s - 0.00 MiB 0.00 MiB 1 2 0 SKIP 0.000667 ms SKIP 79.503748 mW SKIP 53029.000000 pJ
152 slice/negative_axis resize/with_sizes PASS 0.060 s - 0.00 MiB 0.00 MiB 1 3 0 SKIP 0.000797 ms SKIP 92.542033 mW SKIP 73756.000000 pJ
153 slice/negative_indices sigmoid/4d PASS 0.053 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000521 ms SKIP 78.184261 mW SKIP 40734.000000 pJ
154 slice/step2 sigmoid/after_gemm PASS 0.058 s - 0.00 MiB 0.01 MiB 0.00 MiB 0.01 MiB 1 5 0 4 SKIP 0.007513 ms SKIP 105.158653 mW SKIP 790056.960000 pJ
155 softmax/3d_last_axis sigmoid/basic PASS 0.050 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000221 ms SKIP 78.217195 mW SKIP 17286.000000 pJ
156 softmax/basic slice/2d_basic PASS 0.061 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000242 ms SKIP 78.297521 mW SKIP 18948.000000 pJ
157 softmax/channel_axis slice/after_conv PASS 0.058 s - 0.00 MiB 0.00 MiB 0.01 MiB 1 7 0 6 SKIP 0.011296 ms SKIP 118.190765 mW SKIP 1335082.880000 pJ
158 softmax/large_dimension_1024 slice/default_axes PASS 0.061 s - 0.01 MiB 0.00 MiB 0.01 MiB 0.00 MiB 1 0 SKIP 0.000242 ms SKIP 78.297521 mW SKIP 18948.000000 pJ
159 softmax/negative_axis slice/large_channel_1024 PASS 0.058 s - 0.00 MiB 0.01 MiB 0.00 MiB 1 0 SKIP 0.002832 ms SKIP 78.144068 mW SKIP 221304.000000 pJ
160 split/basic slice/nchw_spatial_crop PASS 0.059 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.001302 ms SKIP 78.239631 mW SKIP 101868.000000 pJ
161 split/equal_three_way slice/negative_axis PASS 0.063 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000562 ms SKIP 78.298932 mW SKIP 44004.000000 pJ
162 split/negative_axis slice/negative_indices PASS 0.059 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000322 ms SKIP 78.298137 mW SKIP 25212.000000 pJ
163 split/uneven_channel_axis_4d slice/step2 PASS 0.061 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.002042 ms SKIP 78.293830 mW SKIP 159876.000000 pJ
164 sub/after_gemm softmax/3d_last_axis PASS 0.064 s - 0.01 MiB 0.00 MiB 0.01 MiB 0.00 MiB 5 1 4 0 SKIP UNSUPPORTED SKIP UNSUPPORTED SKIP UNSUPPORTED
165 sub/basic softmax/basic PASS 0.054 s - 0.00 MiB 0.00 MiB 1 0 SKIP UNSUPPORTED SKIP UNSUPPORTED SKIP UNSUPPORTED
166 sub/broadcast_row softmax/channel_axis PASS 0.064 s - 0.00 MiB 0.00 MiB 1 0 SKIP UNSUPPORTED SKIP UNSUPPORTED SKIP UNSUPPORTED
167 sub/channel_broadcast_1024 softmax/large_dimension_1024 PASS 0.063 s - 0.02 MiB 0.01 MiB 0.01 MiB 1 0 SKIP UNSUPPORTED SKIP UNSUPPORTED SKIP UNSUPPORTED
168 sub/constant_lhs_broadcast softmax/negative_axis PASS 0.054 s - 0.00 MiB 0.00 MiB 1 0 SKIP UNSUPPORTED SKIP UNSUPPORTED SKIP UNSUPPORTED
169 sub/leading_dimension_broadcast split/basic PASS 0.054 s - 0.00 MiB 0.00 MiB 1 0 SKIP 0.000403 ms SKIP 78.297767 mW SKIP 31554.000000 pJ
170 split/equal_three_way PASS - 0.00 MiB 0.00 MiB 1 0 0.000564 ms 78.297872 mW 44160.000000 pJ
171 split/negative_axis PASS - 0.00 MiB 0.00 MiB 1 0 0.001083 ms 78.288089 mW 84786.000000 pJ
172 split/uneven_channel_axis_4d PASS - 0.00 MiB 0.00 MiB 1 0 0.000242 ms 78.297521 mW 18948.000000 pJ
173 sub/after_gemm PASS - 0.01 MiB 0.01 MiB 5 4 0.007784 ms 104.703618 mW 815012.960000 pJ
174 sub/basic PASS - 0.00 MiB 0.00 MiB 1 0 0.000323 ms 78.222910 mW 25266.000000 pJ
175 sub/broadcast_row PASS - 0.00 MiB 0.00 MiB 1 0 0.000323 ms 78.222910 mW 25266.000000 pJ
176 sub/channel_broadcast_1024 PASS - 0.02 MiB 0.01 MiB 1 0 0.006913 ms 78.118038 mW 540030.000000 pJ
177 sub/constant_lhs_broadcast PASS - 0.00 MiB 0.00 MiB 1 0 0.000322 ms 78.223602 mW 25188.000000 pJ
178 sub/leading_dimension_broadcast PASS - 0.00 MiB 0.00 MiB 1 0 0.000323 ms 78.222910 mW 25266.000000 pJ