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

update plots
Validate Operations / validate-operations (push) Has been cancelled
Details

Browse Source
This commit is contained in:
NiccoloN
2026-10-26 12:18:36 +01:00
parent 4b53f6cf10
commit 49d6fb42cf
8 changed files with 918 additions and 370 deletions
Download Patch File Download Diff File Expand all files Collapse all files
validation/operations/validation_results.csv
+178
View File
@@ -0,0 +1,178 @@
Operation,Arch,Result (l),Result (t),Compile (l),Host mem (l),Cores mem (l),Cores (l),Xbars (l),Latency (l),Power (l),Energy (l),Compile (t),Host mem (t),Cores mem (t),Cores (t),Xbars (t),Avg latency (t),Throughput (t),Avg power (t),Avg energy (t)
add/after_gemm,arch-a,PASS,PASS,0.040 s,0.01 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.041 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
add/basic,arch-a,PASS,PASS,0.033 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.034 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
add/broadcast_row,arch-a,PASS,PASS,0.034 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.034 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
add/channel_broadcast_1024,arch-a,PASS,PASS,0.035 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,0.035 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,SKIP
add/leading_dimension_broadcast,arch-a,PASS,PASS,0.035 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.035 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
concat/channel_axis,arch-a,PASS,PASS,0.035 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.035 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
concat/negative_axis,arch-a,PASS,PASS,0.033 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.035 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
concat/three_inputs_channel_axis,arch-a,PASS,PASS,0.036 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.035 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
conv/batch_2,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP,0.050 s,0.00 MiB,0.01 MiB,4,2,SKIP,SKIP,SKIP,SKIP
conv/batch_4_pointwise,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.052 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
conv/depthwise_1024_channels,arch-a,PASS,PASS,0.068 s,0.19 MiB,0.38 MiB,128,128,SKIP,SKIP,SKIP,0.116 s,0.26 MiB,0.50 MiB,130,128,SKIP,SKIP,SKIP,SKIP
conv/depthwise_grouped,arch-a,PASS,PASS,0.039 s,0.01 MiB,0.00 MiB,4,4,SKIP,SKIP,SKIP,0.044 s,0.01 MiB,0.00 MiB,6,4,SKIP,SKIP,SKIP,SKIP
conv/dilated_3x3,arch-a,PASS,PASS,0.050 s,0.01 MiB,0.01 MiB,9,9,SKIP,SKIP,SKIP,0.050 s,0.01 MiB,0.01 MiB,10,9,SKIP,SKIP,SKIP,SKIP
conv/dynamic,arch-a,PASS,PASS,0.039 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,0.045 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
conv/explicit_padding,arch-a,PASS,PASS,0.042 s,0.01 MiB,0.02 MiB,16,16,SKIP,SKIP,SKIP,0.055 s,0.01 MiB,0.02 MiB,17,16,SKIP,SKIP,SKIP,SKIP
conv/grouped_many_groups,arch-a,PASS,PASS,0.547 s,0.05 MiB,0.09 MiB,64,64,SKIP,SKIP,SKIP,0.628 s,0.05 MiB,0.14 MiB,124,64,SKIP,SKIP,SKIP,SKIP
conv/grouped_two_groups,arch-a,PASS,PASS,0.049 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP,0.053 s,0.00 MiB,0.00 MiB,4,2,SKIP,SKIP,SKIP,SKIP
conv/huge_pointwise_1024,arch-a,PASS,PASS,0.160 s,0.01 MiB,0.09 MiB,64,64,SKIP,SKIP,SKIP,0.190 s,0.02 MiB,0.11 MiB,74,64,SKIP,SKIP,SKIP,SKIP
conv/huge_pointwise_1024_dynamic,arch-a,PASS,PASS,0.073 s,8.04 MiB,12.55 MiB,167,0,SKIP,SKIP,SKIP,0.249 s,8.05 MiB,12.45 MiB,165,0,SKIP,SKIP,SKIP,SKIP
conv/input_224_7x7_stride2,arch-a,PASS,PASS,0.777 s,24.14 MiB,61.87 MiB,168,169,SKIP,SKIP,SKIP,3.669 s,31.18 MiB,68.93 MiB,165,169,SKIP,SKIP,SKIP,SKIP
conv/kernel_2x2,arch-a,PASS,PASS,0.041 s,0.00 MiB,0.00 MiB,1,1,SKIP,SKIP,SKIP,0.063 s,0.00 MiB,0.00 MiB,3,1,SKIP,SKIP,SKIP,SKIP
conv/kernel_3x3,arch-a,PASS,PASS,0.043 s,0.01 MiB,0.01 MiB,9,9,SKIP,SKIP,SKIP,0.046 s,0.01 MiB,0.01 MiB,11,9,SKIP,SKIP,SKIP,SKIP
conv/kernel_equals_input_spatial,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP,0.048 s,0.00 MiB,0.00 MiB,4,2,SKIP,SKIP,SKIP,SKIP
conv/large_input_channels_1x1,arch-a,PASS,PASS,0.079 s,0.01 MiB,0.01 MiB,8,8,SKIP,SKIP,SKIP,0.081 s,0.01 MiB,0.02 MiB,10,8,SKIP,SKIP,SKIP,SKIP
conv/large_output_channels_1x1,arch-a,PASS,PASS,0.081 s,0.00 MiB,0.01 MiB,8,8,SKIP,SKIP,SKIP,0.087 s,0.01 MiB,0.02 MiB,9,8,SKIP,SKIP,SKIP,SKIP
conv/large_spatial,arch-a,PASS,PASS,0.048 s,0.01 MiB,0.04 MiB,36,36,SKIP,SKIP,SKIP,0.069 s,0.02 MiB,0.04 MiB,37,36,SKIP,SKIP,SKIP,SKIP
conv/multi_channel,arch-a,PASS,PASS,0.043 s,0.00 MiB,0.00 MiB,3,3,SKIP,SKIP,SKIP,0.046 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP,SKIP
conv/non_square_kernel_1x3,arch-a,PASS,PASS,0.045 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP,0.047 s,0.00 MiB,0.00 MiB,3,2,SKIP,SKIP,SKIP,SKIP
conv/non_square_kernel_3x1,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP,0.046 s,0.00 MiB,0.00 MiB,3,2,SKIP,SKIP,SKIP,SKIP
conv/non_uniform_stride,arch-a,PASS,PASS,0.045 s,0.00 MiB,0.00 MiB,3,3,SKIP,SKIP,SKIP,0.046 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP,SKIP
conv/output_channel_grouping_minimal,arch-a,PASS,PASS,0.065 s,0.10 MiB,0.28 MiB,128,128,SKIP,SKIP,SKIP,0.104 s,0.10 MiB,0.25 MiB,128,128,SKIP,SKIP,SKIP,SKIP
conv/pointwise_1x1,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,1,1,SKIP,SKIP,SKIP,0.044 s,0.00 MiB,0.00 MiB,2,1,SKIP,SKIP,SKIP,SKIP
conv/pointwise_tiled_chain,arch-a,PASS,PASS,0.738 s,0.01 MiB,0.03 MiB,17,80,SKIP,SKIP,SKIP,0.745 s,0.02 MiB,0.08 MiB,22,80,SKIP,SKIP,SKIP,SKIP
conv/real_asymmetric_padding,arch-a,PASS,PASS,0.044 s,0.01 MiB,0.03 MiB,28,28,SKIP,SKIP,SKIP,0.060 s,0.01 MiB,0.03 MiB,29,28,SKIP,SKIP,SKIP,SKIP
conv/relu_conv_store,arch-a,PASS,PASS,0.070 s,0.16 MiB,0.66 MiB,168,168,SKIP,SKIP,SKIP,0.120 s,0.17 MiB,0.66 MiB,166,166,SKIP,SKIP,SKIP,SKIP
conv/same_lower_3x3,arch-a,PASS,PASS,0.043 s,0.01 MiB,0.02 MiB,25,25,SKIP,SKIP,SKIP,0.053 s,0.01 MiB,0.02 MiB,26,25,SKIP,SKIP,SKIP,SKIP
conv/same_padding_3x3,arch-a,PASS,PASS,0.046 s,0.01 MiB,0.02 MiB,25,25,SKIP,SKIP,SKIP,0.056 s,0.01 MiB,0.02 MiB,26,25,SKIP,SKIP,SKIP,SKIP
conv/strategy_depthwise_16,arch-a,PASS,PASS,0.062 s,0.06 MiB,0.33 MiB,168,168,SKIP,SKIP,SKIP,0.125 s,0.08 MiB,0.11 MiB,27,52,SKIP,SKIP,SKIP,SKIP
conv/strategy_input_k_tiled,arch-a,PASS,PASS,0.061 s,0.08 MiB,0.27 MiB,108,108,SKIP,SKIP,SKIP,0.091 s,0.08 MiB,0.32 MiB,110,108,SKIP,SKIP,SKIP,SKIP
conv/strategy_output_channel_tiled,arch-a,PASS,PASS,0.062 s,0.03 MiB,0.15 MiB,72,72,SKIP,SKIP,SKIP,0.078 s,0.04 MiB,0.13 MiB,73,72,SKIP,SKIP,SKIP,SKIP
conv/strategy_streamed_packed,arch-a,PASS,PASS,0.131 s,3.34 MiB,7.89 MiB,168,168,SKIP,SKIP,SKIP,0.418 s,3.35 MiB,7.89 MiB,165,165,SKIP,SKIP,SKIP,SKIP
conv/strategy_streamed_patch,arch-a,PASS,PASS,0.074 s,0.34 MiB,1.32 MiB,168,168,SKIP,SKIP,SKIP,0.159 s,0.35 MiB,1.32 MiB,165,165,SKIP,SKIP,SKIP,SKIP
conv/strategy_tiled_2d,arch-a,PASS,PASS,0.096 s,0.10 MiB,0.37 MiB,168,168,SKIP,SKIP,SKIP,0.179 s,0.11 MiB,0.54 MiB,165,168,SKIP,SKIP,SKIP,SKIP
conv/stride_2,arch-a,PASS,PASS,0.053 s,0.01 MiB,0.00 MiB,4,4,SKIP,SKIP,SKIP,0.053 s,0.01 MiB,0.00 MiB,6,4,SKIP,SKIP,SKIP,SKIP
conv/with_bias_3x3,arch-a,PASS,PASS,0.045 s,0.00 MiB,0.01 MiB,3,3,SKIP,SKIP,SKIP,0.047 s,0.00 MiB,0.01 MiB,4,3,SKIP,SKIP,SKIP,SKIP
conv/with_constant,arch-a,PASS,PASS,0.057 s,0.00 MiB,0.00 MiB,1,1,SKIP,SKIP,SKIP,0.069 s,0.00 MiB,0.00 MiB,3,1,SKIP,SKIP,SKIP,SKIP
conv/without_kernel_shape_attr,arch-a,PASS,PASS,0.051 s,0.01 MiB,0.01 MiB,9,9,SKIP,SKIP,SKIP,0.057 s,0.01 MiB,0.01 MiB,11,9,SKIP,SKIP,SKIP,SKIP
conv/yolo11n_depthwise_head,arch-a,PASS,PASS,2.129 s,8.66 MiB,33.48 MiB,164,250,SKIP,SKIP,SKIP,9.599 s,10.25 MiB,36.90 MiB,166,255,SKIP,SKIP,SKIP,SKIP
conv/yolo11n_heavy,arch-a,PASS,PASS,0.188 s,4.82 MiB,19.00 MiB,160,800,SKIP,SKIP,SKIP,0.298 s,4.83 MiB,19.10 MiB,161,800,SKIP,SKIP,SKIP,SKIP
conv/yolo11n_stem,arch-a,PASS,PASS,0.873 s,12.85 MiB,31.85 MiB,168,488,SKIP,SKIP,SKIP,1.318 s,12.88 MiB,31.68 MiB,165,485,SKIP,SKIP,SKIP,SKIP
div/after_gemm,arch-a,PASS,PASS,0.046 s,0.01 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.052 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
div/basic,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
div/channel_broadcast_1024,arch-a,PASS,PASS,0.048 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,0.040 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,SKIP
div/leading_dimension_broadcast,arch-a,PASS,PASS,0.053 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
div/runtime_scalar_rhs,arch-a,PASS,PASS,0.043 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,0.050 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,SKIP
div/scalar_constant,arch-a,PASS,PASS,0.047 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.043 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
gather/3d_input_axis1,arch-a,PASS,PASS,0.041 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.045 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
gather/axis0_matrix_indices,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.039 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
gather/axis1,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
gather/negative_axis,arch-a,PASS,PASS,0.041 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
gather/negative_indices,arch-a,PASS,PASS,0.061 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.049 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
gemm/alpha_beta,arch-a,PASS,PASS,0.050 s,0.01 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.048 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
gemm/bias_rank2_broadcast,arch-a,PASS,PASS,0.045 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.062 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
gemm/dynamic,arch-a,PASS,PASS,0.043 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.049 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
gemm/dynamic_alpha,arch-a,PASS,PASS,0.045 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.045 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
gemm/dynamic_beta,arch-a,PASS,PASS,0.057 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.058 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
gemm/dynamic_bias,arch-a,PASS,PASS,0.047 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.053 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
gemm/dynamic_bias_alpha_beta,arch-a,PASS,PASS,0.043 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.046 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
gemm/dynamic_transpose_b,arch-a,PASS,PASS,0.076 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.077 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
gemm/huge_1024,arch-a,PASS,PASS,0.186 s,0.01 MiB,0.09 MiB,64,64,SKIP,SKIP,SKIP,0.190 s,0.02 MiB,0.10 MiB,73,64,SKIP,SKIP,SKIP,SKIP
gemm/large,arch-a,PASS,PASS,0.050 s,0.02 MiB,0.03 MiB,16,16,SKIP,SKIP,SKIP,0.048 s,0.02 MiB,0.03 MiB,17,16,SKIP,SKIP,SKIP,SKIP
gemm/large_k_small_n,arch-a,PASS,PASS,0.113 s,0.01 MiB,0.01 MiB,8,8,SKIP,SKIP,SKIP,0.088 s,0.01 MiB,0.01 MiB,9,8,SKIP,SKIP,SKIP,SKIP
gemm/non_square,arch-a,PASS,PASS,0.046 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.049 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
gemm/scalar_bias,arch-a,PASS,PASS,0.050 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.057 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
gemm/small,arch-a,PASS,PASS,0.047 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP,0.079 s,0.00 MiB,0.00 MiB,4,2,SKIP,SKIP,SKIP,SKIP
gemm/small_k_large_n,arch-a,PASS,PASS,0.113 s,0.00 MiB,0.01 MiB,8,8,SKIP,SKIP,SKIP,0.110 s,0.01 MiB,0.02 MiB,9,8,SKIP,SKIP,SKIP,SKIP
gemm/square_weights,arch-a,PASS,PASS,0.073 s,0.03 MiB,0.08 MiB,40,40,SKIP,SKIP,SKIP,0.084 s,0.03 MiB,0.09 MiB,43,40,SKIP,SKIP,SKIP,SKIP
gemm/transpose_a,arch-a,PASS,PASS,0.055 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.052 s,0.00 MiB,0.01 MiB,6,4,SKIP,SKIP,SKIP,SKIP
gemm/transpose_a_and_b,arch-a,PASS,PASS,0.048 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.053 s,0.00 MiB,0.01 MiB,6,4,SKIP,SKIP,SKIP,SKIP
gemm/transpose_b,arch-a,PASS,PASS,0.065 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.051 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
gemm/transpose_b_with_bias,arch-a,PASS,PASS,0.053 s,0.01 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.060 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
gemm/with_bias,arch-a,PASS,PASS,0.054 s,0.01 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.044 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
gemv/all_constant,arch-a,PASS,PASS,0.047 s,0.00 MiB,0.00 MiB,0,0,SKIP,SKIP,SKIP,0.040 s,0.00 MiB,0.00 MiB,0,0,SKIP,SKIP,SKIP,SKIP
gemv/constant_weight,arch-a,PASS,PASS,0.058 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,0.059 s,0.00 MiB,0.01 MiB,7,4,SKIP,SKIP,SKIP,SKIP
gemv/non_uniform_bias,arch-a,PASS,PASS,0.062 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,0.061 s,0.00 MiB,0.01 MiB,7,4,SKIP,SKIP,SKIP,SKIP
gemv/scalar_bias,arch-a,PASS,PASS,0.061 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,0.072 s,0.00 MiB,0.01 MiB,7,4,SKIP,SKIP,SKIP,SKIP
gemv/uniform_bias,arch-a,PASS,PASS,0.076 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,0.064 s,0.00 MiB,0.01 MiB,7,4,SKIP,SKIP,SKIP,SKIP
matmul/basic,arch-a,PASS,PASS,0.071 s,0.00 MiB,0.00 MiB,2,2,SKIP,SKIP,SKIP,0.072 s,0.00 MiB,0.00 MiB,4,2,SKIP,SKIP,SKIP,SKIP
matmul/batched_3d,arch-a,PASS,PASS,0.066 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.060 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
matmul/batched_3d_dynamic,arch-a,PASS,PASS,0.048 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.042 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,SKIP
matmul/batched_left_constant,arch-a,PASS,PASS,0.048 s,0.00 MiB,0.02 MiB,8,8,SKIP,SKIP,SKIP,0.054 s,0.01 MiB,0.02 MiB,9,8,SKIP,SKIP,SKIP,SKIP
matmul/batched_lhs_broadcast,arch-a,PASS,PASS,0.047 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.056 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
matmul/batched_rhs_broadcast,arch-a,PASS,PASS,0.052 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.053 s,0.00 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
matmul/dynamic,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.054 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
matmul/huge_1024,arch-a,PASS,PASS,0.174 s,0.01 MiB,0.09 MiB,64,64,SKIP,SKIP,SKIP,0.188 s,0.02 MiB,0.10 MiB,73,64,SKIP,SKIP,SKIP,SKIP
matmul/left_constant,arch-a,PASS,PASS,0.051 s,0.00 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.058 s,0.00 MiB,0.01 MiB,6,4,SKIP,SKIP,SKIP,SKIP
matmul/matrix_vector,arch-a,PASS,PASS,0.120 s,0.52 MiB,0.77 MiB,165,170,SKIP,SKIP,SKIP,0.239 s,0.53 MiB,0.77 MiB,165,173,SKIP,SKIP,SKIP,SKIP
matmul/vector_matrix,arch-a,PASS,PASS,0.088 s,0.01 MiB,0.01 MiB,8,8,SKIP,SKIP,SKIP,0.087 s,0.01 MiB,0.01 MiB,9,8,SKIP,SKIP,SKIP,SKIP
matmul/yolo_attention,arch-a,PASS,PASS,0.394 s,1.02 MiB,39.22 MiB,168,0,SKIP,SKIP,SKIP,0.512 s,2.65 MiB,26.70 MiB,43,0,SKIP,SKIP,SKIP,SKIP
mul/after_conv,arch-a,PASS,PASS,0.065 s,0.00 MiB,0.00 MiB,3,3,SKIP,SKIP,SKIP,0.081 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP,SKIP
mul/after_conv_scalar_constant,arch-a,PASS,PASS,0.053 s,0.00 MiB,0.00 MiB,3,3,SKIP,SKIP,SKIP,0.055 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP,SKIP
mul/basic,arch-a,PASS,PASS,0.075 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.079 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
mul/channel_broadcast_1024,arch-a,PASS,PASS,0.038 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,0.039 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,SKIP
mul/leading_dimension_broadcast,arch-a,PASS,PASS,0.038 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.037 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
mul/scalar_constant,arch-a,PASS,PASS,0.038 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.036 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/avg_basic,arch-a,PASS,PASS,0.047 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.045 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/avg_ceil_mode,arch-a,PASS,PASS,0.061 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.056 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/avg_explicit_padding,arch-a,PASS,PASS,0.047 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.051 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/avg_include_pad,arch-a,PASS,PASS,0.049 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.050 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/avg_large_channels,arch-a,PASS,PASS,0.056 s,0.04 MiB,0.02 MiB,1,0,SKIP,SKIP,SKIP,0.059 s,0.04 MiB,0.02 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/avg_non_uniform_stride,arch-a,PASS,PASS,0.050 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.053 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/avg_real_asymmetric_padding,arch-a,PASS,PASS,0.048 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.048 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/max_after_conv,arch-a,PASS,PASS,0.052 s,0.00 MiB,0.00 MiB,4,4,SKIP,SKIP,SKIP,0.054 s,0.00 MiB,0.00 MiB,5,4,SKIP,SKIP,SKIP,SKIP
pool/max_basic,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.039 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/max_ceil_mode,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/max_global_style_kernel_equals_input,arch-a,PASS,PASS,0.050 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.041 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/max_non_square_kernel,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.067 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/max_real_asymmetric_padding,arch-a,PASS,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/max_same_upper,arch-a,PASS,PASS,0.059 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.049 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
pool/max_stride2_multichannel,arch-a,PASS,PASS,0.045 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.062 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/4d_spatial,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.046 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/4d_spatial_keepdims_0,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,3,0,SKIP,SKIP,SKIP,0.045 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/after_conv,arch-a,PASS,PASS,0.069 s,0.00 MiB,0.00 MiB,3,3,SKIP,SKIP,SKIP,0.067 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP,SKIP
reduce_mean/all_axes_keepdims_0,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.048 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/all_axes_keepdims_1,arch-a,PASS,PASS,0.041 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.052 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/basic,arch-a,PASS,PASS,0.048 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.042 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/channel_axis_nchw,arch-a,PASS,PASS,0.053 s,0.03 MiB,0.02 MiB,4,0,SKIP,SKIP,SKIP,0.048 s,0.03 MiB,0.02 MiB,4,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/keepdims_0,arch-a,PASS,PASS,0.046 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.041 s,0.00 MiB,0.00 MiB,5,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/large_dimension_1024,arch-a,PASS,PASS,0.037 s,0.01 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.039 s,0.01 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/legacy_axes_1_2_keepdims_1,arch-a,PASS,PASS,0.041 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP,0.051 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/legacy_axis1_keepdims_0,arch-a,PASS,PASS,0.049 s,0.00 MiB,0.00 MiB,8,0,SKIP,SKIP,SKIP,0.043 s,0.00 MiB,0.00 MiB,9,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/legacy_axis1_keepdims_1,arch-a,PASS,PASS,0.043 s,0.00 MiB,0.00 MiB,8,0,SKIP,SKIP,SKIP,0.041 s,0.00 MiB,0.00 MiB,8,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/legacy_empty_axes_noop,arch-a,PASS,PASS,0.035 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/legacy_nchw_spatial,arch-a,PASS,PASS,0.048 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.055 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/legacy_negative_axis,arch-a,PASS,PASS,0.050 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP,0.064 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/legacy_reduce_all_keepdims_1,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.041 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reduce_mean/negative_axis,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP,0.046 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP,SKIP
relu/4d,arch-a,PASS,PASS,0.043 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.039 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
relu/after_conv,arch-a,PASS,PASS,0.061 s,0.00 MiB,0.00 MiB,3,3,SKIP,SKIP,SKIP,0.062 s,0.00 MiB,0.00 MiB,4,3,SKIP,SKIP,SKIP,SKIP
relu/after_gemm,arch-a,PASS,PASS,0.044 s,0.01 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.045 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
relu/basic,arch-a,PASS,PASS,0.039 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.038 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reshape/4d_to_2d_flatten,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.045 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reshape/infer_dim_minus_one,arch-a,PASS,PASS,0.041 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reshape/same_rank,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.039 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
reshape/zero_copies_input_dim,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.036 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
resize/height_only,arch-a,PASS,PASS,0.047 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.047 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,SKIP
resize/nearest_2x,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,0.051 s,0.00 MiB,0.00 MiB,4,0,SKIP,SKIP,SKIP,SKIP
resize/nearest_downsample,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP,0.045 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP,SKIP
resize/non_uniform_scales,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP,0.046 s,0.00 MiB,0.00 MiB,6,0,SKIP,SKIP,SKIP,SKIP
resize/width_only,arch-a,PASS,PASS,0.043 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP,0.043 s,0.00 MiB,0.00 MiB,2,0,SKIP,SKIP,SKIP,SKIP
resize/with_sizes,arch-a,PASS,PASS,0.045 s,0.00 MiB,0.00 MiB,3,0,SKIP,SKIP,SKIP,0.047 s,0.00 MiB,0.00 MiB,3,0,SKIP,SKIP,SKIP,SKIP
sigmoid/4d,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
sigmoid/after_gemm,arch-a,PASS,PASS,0.048 s,0.01 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.051 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
sigmoid/basic,arch-a,PASS,PASS,0.038 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.053 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
slice/2d_basic,arch-a,PASS,PASS,0.039 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.043 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
slice/after_conv,arch-a,PASS,PASS,0.050 s,0.00 MiB,0.01 MiB,6,6,SKIP,SKIP,SKIP,0.059 s,0.00 MiB,0.01 MiB,7,6,SKIP,SKIP,SKIP,SKIP
slice/default_axes,arch-a,PASS,PASS,0.039 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.038 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
slice/large_channel_1024,arch-a,PASS,PASS,0.041 s,0.01 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.039 s,0.01 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
slice/nchw_spatial_crop,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
slice/negative_axis,arch-a,PASS,PASS,0.055 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
slice/negative_indices,arch-a,PASS,PASS,0.043 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.038 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
slice/step2,arch-a,PASS,PASS,0.041 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
softmax/3d_last_axis,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
softmax/basic,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.038 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
softmax/channel_axis,arch-a,PASS,PASS,0.037 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.047 s,0.00 MiB,0.00 MiB,3,0,SKIP,SKIP,SKIP,SKIP
softmax/large_dimension_1024,arch-a,PASS,PASS,0.036 s,0.01 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,0.038 s,0.01 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,SKIP
softmax/negative_axis,arch-a,PASS,PASS,0.069 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.060 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
split/basic,arch-a,PASS,PASS,0.052 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.057 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
split/equal_three_way,arch-a,PASS,PASS,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.042 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
split/negative_axis,arch-a,PASS,PASS,0.058 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.043 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
split/uneven_channel_axis_4d,arch-a,PASS,PASS,0.048 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
sub/after_gemm,arch-a,PASS,PASS,0.050 s,0.01 MiB,0.01 MiB,4,4,SKIP,SKIP,SKIP,0.047 s,0.01 MiB,0.01 MiB,5,4,SKIP,SKIP,SKIP,SKIP
sub/basic,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.043 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
sub/broadcast_row,arch-a,PASS,PASS,0.056 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
sub/channel_broadcast_1024,arch-a,PASS,PASS,0.052 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,0.043 s,0.02 MiB,0.01 MiB,1,0,SKIP,SKIP,SKIP,SKIP
sub/constant_lhs_broadcast,arch-a,PASS,PASS,0.044 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.038 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
sub/leading_dimension_broadcast,arch-a,PASS,PASS,0.040 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,0.033 s,0.00 MiB,0.00 MiB,1,0,SKIP,SKIP,SKIP,SKIP
1 Operation Arch Result (l) Result (t) Compile (l) Host mem (l) Cores mem (l) Cores (l) Xbars (l) Latency (l) Power (l) Energy (l) Compile (t) Host mem (t) Cores mem (t) Cores (t) Xbars (t) Avg latency (t) Throughput (t) Avg power (t) Avg energy (t)
2 add/after_gemm arch-a PASS PASS 0.040 s 0.01 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.041 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
3 add/basic arch-a PASS PASS 0.033 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.034 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
4 add/broadcast_row arch-a PASS PASS 0.034 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.034 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
5 add/channel_broadcast_1024 arch-a PASS PASS 0.035 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP 0.035 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP SKIP
6 add/leading_dimension_broadcast arch-a PASS PASS 0.035 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.035 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
7 concat/channel_axis arch-a PASS PASS 0.035 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.035 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
8 concat/negative_axis arch-a PASS PASS 0.033 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.035 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
9 concat/three_inputs_channel_axis arch-a PASS PASS 0.036 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.035 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
10 conv/batch_2 arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 2 2 SKIP SKIP SKIP 0.050 s 0.00 MiB 0.01 MiB 4 2 SKIP SKIP SKIP SKIP
11 conv/batch_4_pointwise arch-a PASS PASS 0.044 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.052 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
12 conv/depthwise_1024_channels arch-a PASS PASS 0.068 s 0.19 MiB 0.38 MiB 128 128 SKIP SKIP SKIP 0.116 s 0.26 MiB 0.50 MiB 130 128 SKIP SKIP SKIP SKIP
13 conv/depthwise_grouped arch-a PASS PASS 0.039 s 0.01 MiB 0.00 MiB 4 4 SKIP SKIP SKIP 0.044 s 0.01 MiB 0.00 MiB 6 4 SKIP SKIP SKIP SKIP
14 conv/dilated_3x3 arch-a PASS PASS 0.050 s 0.01 MiB 0.01 MiB 9 9 SKIP SKIP SKIP 0.050 s 0.01 MiB 0.01 MiB 10 9 SKIP SKIP SKIP SKIP
15 conv/dynamic arch-a PASS PASS 0.039 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP 0.045 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
16 conv/explicit_padding arch-a PASS PASS 0.042 s 0.01 MiB 0.02 MiB 16 16 SKIP SKIP SKIP 0.055 s 0.01 MiB 0.02 MiB 17 16 SKIP SKIP SKIP SKIP
17 conv/grouped_many_groups arch-a PASS PASS 0.547 s 0.05 MiB 0.09 MiB 64 64 SKIP SKIP SKIP 0.628 s 0.05 MiB 0.14 MiB 124 64 SKIP SKIP SKIP SKIP
18 conv/grouped_two_groups arch-a PASS PASS 0.049 s 0.00 MiB 0.00 MiB 2 2 SKIP SKIP SKIP 0.053 s 0.00 MiB 0.00 MiB 4 2 SKIP SKIP SKIP SKIP
19 conv/huge_pointwise_1024 arch-a PASS PASS 0.160 s 0.01 MiB 0.09 MiB 64 64 SKIP SKIP SKIP 0.190 s 0.02 MiB 0.11 MiB 74 64 SKIP SKIP SKIP SKIP
20 conv/huge_pointwise_1024_dynamic arch-a PASS PASS 0.073 s 8.04 MiB 12.55 MiB 167 0 SKIP SKIP SKIP 0.249 s 8.05 MiB 12.45 MiB 165 0 SKIP SKIP SKIP SKIP
21 conv/input_224_7x7_stride2 arch-a PASS PASS 0.777 s 24.14 MiB 61.87 MiB 168 169 SKIP SKIP SKIP 3.669 s 31.18 MiB 68.93 MiB 165 169 SKIP SKIP SKIP SKIP
22 conv/kernel_2x2 arch-a PASS PASS 0.041 s 0.00 MiB 0.00 MiB 1 1 SKIP SKIP SKIP 0.063 s 0.00 MiB 0.00 MiB 3 1 SKIP SKIP SKIP SKIP
23 conv/kernel_3x3 arch-a PASS PASS 0.043 s 0.01 MiB 0.01 MiB 9 9 SKIP SKIP SKIP 0.046 s 0.01 MiB 0.01 MiB 11 9 SKIP SKIP SKIP SKIP
24 conv/kernel_equals_input_spatial arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 2 2 SKIP SKIP SKIP 0.048 s 0.00 MiB 0.00 MiB 4 2 SKIP SKIP SKIP SKIP
25 conv/large_input_channels_1x1 arch-a PASS PASS 0.079 s 0.01 MiB 0.01 MiB 8 8 SKIP SKIP SKIP 0.081 s 0.01 MiB 0.02 MiB 10 8 SKIP SKIP SKIP SKIP
26 conv/large_output_channels_1x1 arch-a PASS PASS 0.081 s 0.00 MiB 0.01 MiB 8 8 SKIP SKIP SKIP 0.087 s 0.01 MiB 0.02 MiB 9 8 SKIP SKIP SKIP SKIP
27 conv/large_spatial arch-a PASS PASS 0.048 s 0.01 MiB 0.04 MiB 36 36 SKIP SKIP SKIP 0.069 s 0.02 MiB 0.04 MiB 37 36 SKIP SKIP SKIP SKIP
28 conv/multi_channel arch-a PASS PASS 0.043 s 0.00 MiB 0.00 MiB 3 3 SKIP SKIP SKIP 0.046 s 0.00 MiB 0.00 MiB 4 3 SKIP SKIP SKIP SKIP
29 conv/non_square_kernel_1x3 arch-a PASS PASS 0.045 s 0.00 MiB 0.00 MiB 2 2 SKIP SKIP SKIP 0.047 s 0.00 MiB 0.00 MiB 3 2 SKIP SKIP SKIP SKIP
30 conv/non_square_kernel_3x1 arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 2 2 SKIP SKIP SKIP 0.046 s 0.00 MiB 0.00 MiB 3 2 SKIP SKIP SKIP SKIP
31 conv/non_uniform_stride arch-a PASS PASS 0.045 s 0.00 MiB 0.00 MiB 3 3 SKIP SKIP SKIP 0.046 s 0.00 MiB 0.00 MiB 4 3 SKIP SKIP SKIP SKIP
32 conv/output_channel_grouping_minimal arch-a PASS PASS 0.065 s 0.10 MiB 0.28 MiB 128 128 SKIP SKIP SKIP 0.104 s 0.10 MiB 0.25 MiB 128 128 SKIP SKIP SKIP SKIP
33 conv/pointwise_1x1 arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 1 1 SKIP SKIP SKIP 0.044 s 0.00 MiB 0.00 MiB 2 1 SKIP SKIP SKIP SKIP
34 conv/pointwise_tiled_chain arch-a PASS PASS 0.738 s 0.01 MiB 0.03 MiB 17 80 SKIP SKIP SKIP 0.745 s 0.02 MiB 0.08 MiB 22 80 SKIP SKIP SKIP SKIP
35 conv/real_asymmetric_padding arch-a PASS PASS 0.044 s 0.01 MiB 0.03 MiB 28 28 SKIP SKIP SKIP 0.060 s 0.01 MiB 0.03 MiB 29 28 SKIP SKIP SKIP SKIP
36 conv/relu_conv_store arch-a PASS PASS 0.070 s 0.16 MiB 0.66 MiB 168 168 SKIP SKIP SKIP 0.120 s 0.17 MiB 0.66 MiB 166 166 SKIP SKIP SKIP SKIP
37 conv/same_lower_3x3 arch-a PASS PASS 0.043 s 0.01 MiB 0.02 MiB 25 25 SKIP SKIP SKIP 0.053 s 0.01 MiB 0.02 MiB 26 25 SKIP SKIP SKIP SKIP
38 conv/same_padding_3x3 arch-a PASS PASS 0.046 s 0.01 MiB 0.02 MiB 25 25 SKIP SKIP SKIP 0.056 s 0.01 MiB 0.02 MiB 26 25 SKIP SKIP SKIP SKIP
39 conv/strategy_depthwise_16 arch-a PASS PASS 0.062 s 0.06 MiB 0.33 MiB 168 168 SKIP SKIP SKIP 0.125 s 0.08 MiB 0.11 MiB 27 52 SKIP SKIP SKIP SKIP
40 conv/strategy_input_k_tiled arch-a PASS PASS 0.061 s 0.08 MiB 0.27 MiB 108 108 SKIP SKIP SKIP 0.091 s 0.08 MiB 0.32 MiB 110 108 SKIP SKIP SKIP SKIP
41 conv/strategy_output_channel_tiled arch-a PASS PASS 0.062 s 0.03 MiB 0.15 MiB 72 72 SKIP SKIP SKIP 0.078 s 0.04 MiB 0.13 MiB 73 72 SKIP SKIP SKIP SKIP
42 conv/strategy_streamed_packed arch-a PASS PASS 0.131 s 3.34 MiB 7.89 MiB 168 168 SKIP SKIP SKIP 0.418 s 3.35 MiB 7.89 MiB 165 165 SKIP SKIP SKIP SKIP
43 conv/strategy_streamed_patch arch-a PASS PASS 0.074 s 0.34 MiB 1.32 MiB 168 168 SKIP SKIP SKIP 0.159 s 0.35 MiB 1.32 MiB 165 165 SKIP SKIP SKIP SKIP
44 conv/strategy_tiled_2d arch-a PASS PASS 0.096 s 0.10 MiB 0.37 MiB 168 168 SKIP SKIP SKIP 0.179 s 0.11 MiB 0.54 MiB 165 168 SKIP SKIP SKIP SKIP
45 conv/stride_2 arch-a PASS PASS 0.053 s 0.01 MiB 0.00 MiB 4 4 SKIP SKIP SKIP 0.053 s 0.01 MiB 0.00 MiB 6 4 SKIP SKIP SKIP SKIP
46 conv/with_bias_3x3 arch-a PASS PASS 0.045 s 0.00 MiB 0.01 MiB 3 3 SKIP SKIP SKIP 0.047 s 0.00 MiB 0.01 MiB 4 3 SKIP SKIP SKIP SKIP
47 conv/with_constant arch-a PASS PASS 0.057 s 0.00 MiB 0.00 MiB 1 1 SKIP SKIP SKIP 0.069 s 0.00 MiB 0.00 MiB 3 1 SKIP SKIP SKIP SKIP
48 conv/without_kernel_shape_attr arch-a PASS PASS 0.051 s 0.01 MiB 0.01 MiB 9 9 SKIP SKIP SKIP 0.057 s 0.01 MiB 0.01 MiB 11 9 SKIP SKIP SKIP SKIP
49 conv/yolo11n_depthwise_head arch-a PASS PASS 2.129 s 8.66 MiB 33.48 MiB 164 250 SKIP SKIP SKIP 9.599 s 10.25 MiB 36.90 MiB 166 255 SKIP SKIP SKIP SKIP
50 conv/yolo11n_heavy arch-a PASS PASS 0.188 s 4.82 MiB 19.00 MiB 160 800 SKIP SKIP SKIP 0.298 s 4.83 MiB 19.10 MiB 161 800 SKIP SKIP SKIP SKIP
51 conv/yolo11n_stem arch-a PASS PASS 0.873 s 12.85 MiB 31.85 MiB 168 488 SKIP SKIP SKIP 1.318 s 12.88 MiB 31.68 MiB 165 485 SKIP SKIP SKIP SKIP
52 div/after_gemm arch-a PASS PASS 0.046 s 0.01 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.052 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
53 div/basic arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
54 div/channel_broadcast_1024 arch-a PASS PASS 0.048 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP 0.040 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP SKIP
55 div/leading_dimension_broadcast arch-a PASS PASS 0.053 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.058 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
56 div/runtime_scalar_rhs arch-a PASS PASS 0.043 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP 0.050 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP SKIP
57 div/scalar_constant arch-a PASS PASS 0.047 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.043 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
58 gather/3d_input_axis1 arch-a PASS PASS 0.041 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.045 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
59 gather/axis0_matrix_indices arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.039 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
60 gather/axis1 arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
61 gather/negative_axis arch-a PASS PASS 0.041 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
62 gather/negative_indices arch-a PASS PASS 0.061 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.049 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
63 gemm/alpha_beta arch-a PASS PASS 0.050 s 0.01 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.048 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
64 gemm/bias_rank2_broadcast arch-a PASS PASS 0.045 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.062 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
65 gemm/dynamic arch-a PASS PASS 0.043 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.049 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
66 gemm/dynamic_alpha arch-a PASS PASS 0.045 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.045 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
67 gemm/dynamic_beta arch-a PASS PASS 0.057 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.058 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
68 gemm/dynamic_bias arch-a PASS PASS 0.047 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.053 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
69 gemm/dynamic_bias_alpha_beta arch-a PASS PASS 0.043 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.046 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
70 gemm/dynamic_transpose_b arch-a PASS PASS 0.076 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.077 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
71 gemm/huge_1024 arch-a PASS PASS 0.186 s 0.01 MiB 0.09 MiB 64 64 SKIP SKIP SKIP 0.190 s 0.02 MiB 0.10 MiB 73 64 SKIP SKIP SKIP SKIP
72 gemm/large arch-a PASS PASS 0.050 s 0.02 MiB 0.03 MiB 16 16 SKIP SKIP SKIP 0.048 s 0.02 MiB 0.03 MiB 17 16 SKIP SKIP SKIP SKIP
73 gemm/large_k_small_n arch-a PASS PASS 0.113 s 0.01 MiB 0.01 MiB 8 8 SKIP SKIP SKIP 0.088 s 0.01 MiB 0.01 MiB 9 8 SKIP SKIP SKIP SKIP
74 gemm/non_square arch-a PASS PASS 0.046 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.049 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
75 gemm/scalar_bias arch-a PASS PASS 0.050 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.057 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
76 gemm/small arch-a PASS PASS 0.047 s 0.00 MiB 0.00 MiB 2 2 SKIP SKIP SKIP 0.079 s 0.00 MiB 0.00 MiB 4 2 SKIP SKIP SKIP SKIP
77 gemm/small_k_large_n arch-a PASS PASS 0.113 s 0.00 MiB 0.01 MiB 8 8 SKIP SKIP SKIP 0.110 s 0.01 MiB 0.02 MiB 9 8 SKIP SKIP SKIP SKIP
78 gemm/square_weights arch-a PASS PASS 0.073 s 0.03 MiB 0.08 MiB 40 40 SKIP SKIP SKIP 0.084 s 0.03 MiB 0.09 MiB 43 40 SKIP SKIP SKIP SKIP
79 gemm/transpose_a arch-a PASS PASS 0.055 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.052 s 0.00 MiB 0.01 MiB 6 4 SKIP SKIP SKIP SKIP
80 gemm/transpose_a_and_b arch-a PASS PASS 0.048 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.053 s 0.00 MiB 0.01 MiB 6 4 SKIP SKIP SKIP SKIP
81 gemm/transpose_b arch-a PASS PASS 0.065 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.051 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
82 gemm/transpose_b_with_bias arch-a PASS PASS 0.053 s 0.01 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.060 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
83 gemm/with_bias arch-a PASS PASS 0.054 s 0.01 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.044 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
84 gemv/all_constant arch-a PASS PASS 0.047 s 0.00 MiB 0.00 MiB 0 0 SKIP SKIP SKIP 0.040 s 0.00 MiB 0.00 MiB 0 0 SKIP SKIP SKIP SKIP
85 gemv/constant_weight arch-a PASS PASS 0.058 s 0.00 MiB 0.01 MiB 5 4 SKIP SKIP SKIP 0.059 s 0.00 MiB 0.01 MiB 7 4 SKIP SKIP SKIP SKIP
86 gemv/non_uniform_bias arch-a PASS PASS 0.062 s 0.00 MiB 0.01 MiB 5 4 SKIP SKIP SKIP 0.061 s 0.00 MiB 0.01 MiB 7 4 SKIP SKIP SKIP SKIP
87 gemv/scalar_bias arch-a PASS PASS 0.061 s 0.00 MiB 0.01 MiB 5 4 SKIP SKIP SKIP 0.072 s 0.00 MiB 0.01 MiB 7 4 SKIP SKIP SKIP SKIP
88 gemv/uniform_bias arch-a PASS PASS 0.076 s 0.00 MiB 0.01 MiB 5 4 SKIP SKIP SKIP 0.064 s 0.00 MiB 0.01 MiB 7 4 SKIP SKIP SKIP SKIP
89 matmul/basic arch-a PASS PASS 0.071 s 0.00 MiB 0.00 MiB 2 2 SKIP SKIP SKIP 0.072 s 0.00 MiB 0.00 MiB 4 2 SKIP SKIP SKIP SKIP
90 matmul/batched_3d arch-a PASS PASS 0.066 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.060 s 0.00 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
91 matmul/batched_3d_dynamic arch-a PASS PASS 0.048 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.042 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP SKIP
92 matmul/batched_left_constant arch-a PASS PASS 0.048 s 0.00 MiB 0.02 MiB 8 8 SKIP SKIP SKIP 0.054 s 0.01 MiB 0.02 MiB 9 8 SKIP SKIP SKIP SKIP
93 matmul/batched_lhs_broadcast arch-a PASS PASS 0.047 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.056 s 0.00 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
94 matmul/batched_rhs_broadcast arch-a PASS PASS 0.052 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.053 s 0.00 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
95 matmul/dynamic arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.054 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
96 matmul/huge_1024 arch-a PASS PASS 0.174 s 0.01 MiB 0.09 MiB 64 64 SKIP SKIP SKIP 0.188 s 0.02 MiB 0.10 MiB 73 64 SKIP SKIP SKIP SKIP
97 matmul/left_constant arch-a PASS PASS 0.051 s 0.00 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.058 s 0.00 MiB 0.01 MiB 6 4 SKIP SKIP SKIP SKIP
98 matmul/matrix_vector arch-a PASS PASS 0.120 s 0.52 MiB 0.77 MiB 165 170 SKIP SKIP SKIP 0.239 s 0.53 MiB 0.77 MiB 165 173 SKIP SKIP SKIP SKIP
99 matmul/vector_matrix arch-a PASS PASS 0.088 s 0.01 MiB 0.01 MiB 8 8 SKIP SKIP SKIP 0.087 s 0.01 MiB 0.01 MiB 9 8 SKIP SKIP SKIP SKIP
100 matmul/yolo_attention arch-a PASS PASS 0.394 s 1.02 MiB 39.22 MiB 168 0 SKIP SKIP SKIP 0.512 s 2.65 MiB 26.70 MiB 43 0 SKIP SKIP SKIP SKIP
101 mul/after_conv arch-a PASS PASS 0.065 s 0.00 MiB 0.00 MiB 3 3 SKIP SKIP SKIP 0.081 s 0.00 MiB 0.00 MiB 4 3 SKIP SKIP SKIP SKIP
102 mul/after_conv_scalar_constant arch-a PASS PASS 0.053 s 0.00 MiB 0.00 MiB 3 3 SKIP SKIP SKIP 0.055 s 0.00 MiB 0.00 MiB 4 3 SKIP SKIP SKIP SKIP
103 mul/basic arch-a PASS PASS 0.075 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.079 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
104 mul/channel_broadcast_1024 arch-a PASS PASS 0.038 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP 0.039 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP SKIP
105 mul/leading_dimension_broadcast arch-a PASS PASS 0.038 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.037 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
106 mul/scalar_constant arch-a PASS PASS 0.038 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.036 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
107 pool/avg_basic arch-a PASS PASS 0.047 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.045 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
108 pool/avg_ceil_mode arch-a PASS PASS 0.061 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.056 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
109 pool/avg_explicit_padding arch-a PASS PASS 0.047 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.051 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
110 pool/avg_include_pad arch-a PASS PASS 0.049 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.050 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
111 pool/avg_large_channels arch-a PASS PASS 0.056 s 0.04 MiB 0.02 MiB 1 0 SKIP SKIP SKIP 0.059 s 0.04 MiB 0.02 MiB 1 0 SKIP SKIP SKIP SKIP
112 pool/avg_non_uniform_stride arch-a PASS PASS 0.050 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.053 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
113 pool/avg_real_asymmetric_padding arch-a PASS PASS 0.048 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.048 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
114 pool/max_after_conv arch-a PASS PASS 0.052 s 0.00 MiB 0.00 MiB 4 4 SKIP SKIP SKIP 0.054 s 0.00 MiB 0.00 MiB 5 4 SKIP SKIP SKIP SKIP
115 pool/max_basic arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.039 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
116 pool/max_ceil_mode arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
117 pool/max_global_style_kernel_equals_input arch-a PASS PASS 0.050 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.041 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
118 pool/max_non_square_kernel arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.067 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
119 pool/max_real_asymmetric_padding arch-a PASS PASS 0.059 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.059 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
120 pool/max_same_upper arch-a PASS PASS 0.059 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.049 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
121 pool/max_stride2_multichannel arch-a PASS PASS 0.045 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.062 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
122 reduce_mean/4d_spatial arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.046 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
123 reduce_mean/4d_spatial_keepdims_0 arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 3 0 SKIP SKIP SKIP 0.045 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP SKIP
124 reduce_mean/after_conv arch-a PASS PASS 0.069 s 0.00 MiB 0.00 MiB 3 3 SKIP SKIP SKIP 0.067 s 0.00 MiB 0.00 MiB 4 3 SKIP SKIP SKIP SKIP
125 reduce_mean/all_axes_keepdims_0 arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.048 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
126 reduce_mean/all_axes_keepdims_1 arch-a PASS PASS 0.041 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.052 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
127 reduce_mean/basic arch-a PASS PASS 0.048 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.042 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP SKIP
128 reduce_mean/channel_axis_nchw arch-a PASS PASS 0.053 s 0.03 MiB 0.02 MiB 4 0 SKIP SKIP SKIP 0.048 s 0.03 MiB 0.02 MiB 4 0 SKIP SKIP SKIP SKIP
129 reduce_mean/keepdims_0 arch-a PASS PASS 0.046 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.041 s 0.00 MiB 0.00 MiB 5 0 SKIP SKIP SKIP SKIP
130 reduce_mean/large_dimension_1024 arch-a PASS PASS 0.037 s 0.01 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.039 s 0.01 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
131 reduce_mean/legacy_axes_1_2_keepdims_1 arch-a PASS PASS 0.041 s 0.00 MiB 0.00 MiB 2 0 SKIP SKIP SKIP 0.051 s 0.00 MiB 0.00 MiB 2 0 SKIP SKIP SKIP SKIP
132 reduce_mean/legacy_axis1_keepdims_0 arch-a PASS PASS 0.049 s 0.00 MiB 0.00 MiB 8 0 SKIP SKIP SKIP 0.043 s 0.00 MiB 0.00 MiB 9 0 SKIP SKIP SKIP SKIP
133 reduce_mean/legacy_axis1_keepdims_1 arch-a PASS PASS 0.043 s 0.00 MiB 0.00 MiB 8 0 SKIP SKIP SKIP 0.041 s 0.00 MiB 0.00 MiB 8 0 SKIP SKIP SKIP SKIP
134 reduce_mean/legacy_empty_axes_noop arch-a PASS PASS 0.035 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
135 reduce_mean/legacy_nchw_spatial arch-a PASS PASS 0.048 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.055 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
136 reduce_mean/legacy_negative_axis arch-a PASS PASS 0.050 s 0.00 MiB 0.00 MiB 6 0 SKIP SKIP SKIP 0.064 s 0.00 MiB 0.00 MiB 6 0 SKIP SKIP SKIP SKIP
137 reduce_mean/legacy_reduce_all_keepdims_1 arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.041 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
138 reduce_mean/negative_axis arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 6 0 SKIP SKIP SKIP 0.046 s 0.00 MiB 0.00 MiB 6 0 SKIP SKIP SKIP SKIP
139 relu/4d arch-a PASS PASS 0.043 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.039 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
140 relu/after_conv arch-a PASS PASS 0.061 s 0.00 MiB 0.00 MiB 3 3 SKIP SKIP SKIP 0.062 s 0.00 MiB 0.00 MiB 4 3 SKIP SKIP SKIP SKIP
141 relu/after_gemm arch-a PASS PASS 0.044 s 0.01 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.045 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
142 relu/basic arch-a PASS PASS 0.039 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.038 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
143 reshape/4d_to_2d_flatten arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.045 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
144 reshape/infer_dim_minus_one arch-a PASS PASS 0.041 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
145 reshape/same_rank arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.039 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
146 reshape/zero_copies_input_dim arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.036 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
147 resize/height_only arch-a PASS PASS 0.047 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.047 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP SKIP
148 resize/nearest_2x arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP 0.051 s 0.00 MiB 0.00 MiB 4 0 SKIP SKIP SKIP SKIP
149 resize/nearest_downsample arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 2 0 SKIP SKIP SKIP 0.045 s 0.00 MiB 0.00 MiB 2 0 SKIP SKIP SKIP SKIP
150 resize/non_uniform_scales arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 6 0 SKIP SKIP SKIP 0.046 s 0.00 MiB 0.00 MiB 6 0 SKIP SKIP SKIP SKIP
151 resize/width_only arch-a PASS PASS 0.043 s 0.00 MiB 0.00 MiB 2 0 SKIP SKIP SKIP 0.043 s 0.00 MiB 0.00 MiB 2 0 SKIP SKIP SKIP SKIP
152 resize/with_sizes arch-a PASS PASS 0.045 s 0.00 MiB 0.00 MiB 3 0 SKIP SKIP SKIP 0.047 s 0.00 MiB 0.00 MiB 3 0 SKIP SKIP SKIP SKIP
153 sigmoid/4d arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.058 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
154 sigmoid/after_gemm arch-a PASS PASS 0.048 s 0.01 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.051 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
155 sigmoid/basic arch-a PASS PASS 0.038 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.053 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
156 slice/2d_basic arch-a PASS PASS 0.039 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.043 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
157 slice/after_conv arch-a PASS PASS 0.050 s 0.00 MiB 0.01 MiB 6 6 SKIP SKIP SKIP 0.059 s 0.00 MiB 0.01 MiB 7 6 SKIP SKIP SKIP SKIP
158 slice/default_axes arch-a PASS PASS 0.039 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.038 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
159 slice/large_channel_1024 arch-a PASS PASS 0.041 s 0.01 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.039 s 0.01 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
160 slice/nchw_spatial_crop arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
161 slice/negative_axis arch-a PASS PASS 0.055 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
162 slice/negative_indices arch-a PASS PASS 0.043 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.038 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
163 slice/step2 arch-a PASS PASS 0.041 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
164 softmax/3d_last_axis arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
165 softmax/basic arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.038 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
166 softmax/channel_axis arch-a PASS PASS 0.037 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.047 s 0.00 MiB 0.00 MiB 3 0 SKIP SKIP SKIP SKIP
167 softmax/large_dimension_1024 arch-a PASS PASS 0.036 s 0.01 MiB 0.01 MiB 1 0 SKIP SKIP SKIP 0.038 s 0.01 MiB 0.01 MiB 1 0 SKIP SKIP SKIP SKIP
168 softmax/negative_axis arch-a PASS PASS 0.069 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.060 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
169 split/basic arch-a PASS PASS 0.052 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.057 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
170 split/equal_three_way arch-a PASS PASS 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.042 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
171 split/negative_axis arch-a PASS PASS 0.058 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.043 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
172 split/uneven_channel_axis_4d arch-a PASS PASS 0.048 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
173 sub/after_gemm arch-a PASS PASS 0.050 s 0.01 MiB 0.01 MiB 4 4 SKIP SKIP SKIP 0.047 s 0.01 MiB 0.01 MiB 5 4 SKIP SKIP SKIP SKIP
174 sub/basic arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.043 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
175 sub/broadcast_row arch-a PASS PASS 0.056 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
176 sub/channel_broadcast_1024 arch-a PASS PASS 0.052 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP 0.043 s 0.02 MiB 0.01 MiB 1 0 SKIP SKIP SKIP SKIP
177 sub/constant_lhs_broadcast arch-a PASS PASS 0.044 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.038 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP
178 sub/leading_dimension_broadcast arch-a PASS PASS 0.040 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP 0.033 s 0.00 MiB 0.00 MiB 1 0 SKIP SKIP SKIP SKIP