English | 中文版

Appendix E: Complete Kernel Inventory

This appendix is auto-generated by scripts/generate_kernel_appendix.sh. Run bash scripts/generate_kernel_appendix.sh to regenerate.

Summary

Metric	Count
Compiletest kernels	486
Deployable kernels	19
Total kernels	505
MultiKernelBench coverage	300/300 (100%)
MKB categories covered	15/15 (100%)
Memory safety vulnerability patterns	6 classes (with attack examples)

Vulnerability Pattern Legend

ID	Vulnerability	C++ Root Cause	Rust Prevention	Attack Example
V1	Type erasure	`GM_ADDR` erases all type info	Function signature encodes element type	`case1`
V2	Buffer overflow	`GetValue(i)` unchecked indexing	Buffer-ID API with explicit count	`case2`
V3	Integer overflow	Silent `u32` wrap in offset calc	`wrapping_mul` makes overflow explicit	`case6`
V4	Use-after-free	`FreeTensor()` then stale access	No manual free in API	`case3`
V5	Double free	`FreeTensor()` called twice	No free operation exists	`case5`
V6	Missing sync	Forgotten `pipe_barrier()`	`kernel_ops` composites embed barriers	`case4`

Kernel Inventory by Category

Activation (17 kernels)

Applicable vulnerability patterns: V1(type erasure),V2(unchecked index),V6(missing sync)

MKB reference: reference_kernels/activation/

Kernel Function	Source File	MKB Reference	910B3 Status
`abs_kernel`	`tests/compiletest/ui/abs_kernel.rs`	—	PASS
`relu`	`tests/compiletest/ui/relu_kernel.rs`	`relu.py`	PASS
`sigmoid`	`tests/compiletest/ui/sigmoid_kernel.rs`	`sigmoid.py`	PASS
`tanh_kernel`	`tests/compiletest/ui/tanh_kernel.rs`	`tanh_kernel.py`	PASS
`gelu`	`tests/compiletest/ui/gelu_kernel.rs`	`gelu.py`	PASS
`elu`	`tests/compiletest/ui/elu_kernel.rs`	`elu.py`	PASS
`softplus`	`tests/compiletest/ui/softplus_kernel.rs`	`softplus.py`	PASS
`leaky_relu`	`tests/compiletest/ui/leaky_relu_kernel.rs`	`leaky_relu.py`	PASS
`softmax`	`tests/compiletest/ui/softmax_kernel.rs`	`softmax.py`	PASS
`log_softmax`	`tests/compiletest/ui/log_softmax_kernel.rs`	`log_softmax.py`	PASS
`test_selu`	`tests/compiletest/ui/selu_swish_kernel.rs`	`test_selu.py`	PASS
`test_swish`	`tests/compiletest/ui/selu_swish_kernel.rs`	`test_swish.py`	PASS
`softsign`	`tests/compiletest/ui/softsign_kernel.rs`	`softsign.py`	PASS
`hardsigmoid`	`tests/compiletest/ui/hardsigmoid_kernel.rs`	`hardsigmoid.py`	PASS
`hardswish`	`tests/compiletest/ui/hardswish_kernel.rs`	`hardswish.py`	PASS
`mish`	`tests/compiletest/ui/mish_kernel.rs`	`mish.py`	PASS
`gelu_tanh`	`tests/compiletest/ui/gelu_tanh_kernel.rs`	`gelu_tanh.py`	PASS

Architecture (77 kernels)

Applicable vulnerability patterns: V1,V2,V3(offset overflow),V6

MKB reference: reference_kernels/architecture/

Kernel Function	Source File	MKB Reference	910B3 Status
`mlp_relu`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`mlp_gelu_bias`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`mlp_swish`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`ffn_prenorm`	`tests/compiletest/ui/arch_ops_kernel.rs`	`ffn_prenorm.py`	PASS
`down_proj`	`tests/compiletest/ui/arch_ops_kernel.rs`	`down_proj.py`	PASS
`attention_score_norm`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`rope_freq`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`embedding_scale`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`gated_residual`	`tests/compiletest/ui/arch_ops_kernel.rs`	`gated_residual.py`	PASS
`scaled_dot`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`classifier_head`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`regression_head`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`softmax_classifier`	`tests/compiletest/ui/arch_ops_kernel.rs`	—	PASS
`mlp`	`tests/compiletest/ui/arch_ops_kernel.rs`	`mlp.py`	PASS
`deep_narrow_mlp`	`tests/compiletest/ui/arch_ops_kernel.rs`	`deep_narrow_mlp.py`	PASS
`shallow_wide_mlp`	`tests/compiletest/ui/arch_ops_kernel.rs`	`shallow_wide_mlp.py`	PASS
`vanilla_rnn`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`vanilla_rnn.py`	PASS
`lstm_forget_gate`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`lstm_input_gate`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`lstm_cell_candidate`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`lstm_cell_update`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`lstm_output`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`gru_reset_gate`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`gru_update_gate`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`gru_candidate`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`gru_hidden_update`	`tests/compiletest/ui/arch_rnn_kernel.rs`	—	PASS
`vanilla_rnn_hidden`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`vanilla_rnn_hidden.py`	PASS
`lstm`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`lstm.py`	PASS
`lstm_bidirectional`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`lstm_bidirectional.py`	PASS
`lstm_cn`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`lstm_cn.py`	PASS
`gru`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`gru.py`	PASS
`gru_birectional`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`gru_birectional.py`	PASS
`gru_bidirectional_hidden`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`gru_bidirectional_hidden.py`	PASS
`gru_hidden`	`tests/compiletest/ui/arch_rnn_kernel.rs`	`gru_hidden.py`	PASS
`alexnet_fc`	`tests/compiletest/ui/arch_network_kernel.rs`	`alexnet_fc.py`	PASS
`vgg_fc`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`resnet_residual`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`densenet_block`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`mobilenet_pointwise`	`tests/compiletest/ui/arch_network_kernel.rs`	`mobilenet_pointwise.py`	PASS
`efficientnet_fc`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`inception_merge`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`squeezenet_fire`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`shufflenet_fc`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`regnet_stem`	`tests/compiletest/ui/arch_network_kernel.rs`	`regnet_stem.py`	PASS
`lenet_fc`	`tests/compiletest/ui/arch_network_kernel.rs`	`lenet_fc.py`	PASS
`unet_skip`	`tests/compiletest/ui/arch_network_kernel.rs`	`unet_skip.py`	PASS
`vit_mlp`	`tests/compiletest/ui/arch_network_kernel.rs`	`vit_mlp.py`	PASS
`swin_attention`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`mingpt_block`	`tests/compiletest/ui/arch_network_kernel.rs`	`mingpt_block.py`	PASS
`mlp_mixer`	`tests/compiletest/ui/arch_network_kernel.rs`	`mlp_mixer.py`	PASS
`mamba_ssm`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`densenet121`	`tests/compiletest/ui/arch_network_kernel.rs`	`densenet121.py`	PASS
`densenet121_dense_block`	`tests/compiletest/ui/arch_network_kernel.rs`	`densenet121_dense_block.py`	PASS
`densenet121_transition_layer`	`tests/compiletest/ui/arch_network_kernel.rs`	`densenet121_transition_layer.py`	PASS
`densenet201`	`tests/compiletest/ui/arch_network_kernel.rs`	`densenet201.py`	PASS
`efficientnet_b0`	`tests/compiletest/ui/arch_network_kernel.rs`	`efficientnet_b0.py`	PASS
`efficientnet_b1`	`tests/compiletest/ui/arch_network_kernel.rs`	`efficientnet_b1.py`	PASS
`efficientnet_b2`	`tests/compiletest/ui/arch_network_kernel.rs`	`efficientnet_b2.py`	PASS
`resnet18`	`tests/compiletest/ui/arch_network_kernel.rs`	`resnet18.py`	PASS
`resnet101`	`tests/compiletest/ui/arch_network_kernel.rs`	`resnet101.py`	PASS
`resnet_basic_block`	`tests/compiletest/ui/arch_network_kernel.rs`	`resnet_basic_block.py`	PASS
`vgg16`	`tests/compiletest/ui/arch_network_kernel.rs`	`vgg16.py`	PASS
`vgg19`	`tests/compiletest/ui/arch_network_kernel.rs`	`vgg19.py`	PASS
`squeeze_net`	`tests/compiletest/ui/arch_network_kernel.rs`	`squeeze_net.py`	PASS
`squeeze_net_fire_module`	`tests/compiletest/ui/arch_network_kernel.rs`	`squeeze_net_fire_module.py`	PASS
`shufflenet`	`tests/compiletest/ui/arch_network_kernel.rs`	`shufflenet.py`	PASS
`shufflenet_unit`	`tests/compiletest/ui/arch_network_kernel.rs`	`shufflenet_unit.py`	PASS
`googlenet_inception_module`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`googlenet_inception_v1`	`tests/compiletest/ui/arch_network_kernel.rs`	—	PASS
`swin_mlp`	`tests/compiletest/ui/arch_network_kernel.rs`	`swin_mlp.py`	PASS
`swintransformer_v2`	`tests/compiletest/ui/arch_network_kernel.rs`	`swintransformer_v2.py`	PASS
`mamba_return_final_state`	`tests/compiletest/ui/arch_network_kernel.rs`	`mamba_return_final_state.py`	PASS
`mamba_return_y`	`tests/compiletest/ui/arch_network_kernel.rs`	`mamba_return_y.py`	PASS
`convolutional_vision_transformer`	`tests/compiletest/ui/arch_network_kernel.rs`	`convolutional_vision_transformer.py`	PASS
`net_vlad_no_ghost_clusters`	`tests/compiletest/ui/arch_network_kernel.rs`	`net_vlad_no_ghost_clusters.py`	PASS
`net_vlad_with_ghost_clusters`	`tests/compiletest/ui/arch_network_kernel.rs`	`net_vlad_with_ghost_clusters.py`	PASS
`mobilenetv2_inverted`	`tests/compiletest/ui/arch_network_kernel.rs`	`mobilenetv2_inverted.py`	PASS

Attention (23 kernels)

Applicable vulnerability patterns: V1,V2,V3,V6(multi-stage sync)

MKB reference: reference_kernels/attention/

Kernel Function	Source File	MKB Reference	910B3 Status
`attention_softmax`	`tests/compiletest/ui/attention_kernel.rs`	—	PASS
`residual_add_layernorm`	`tests/compiletest/ui/attention_kernel.rs`	—	PASS
`residual_add_rmsnorm`	`tests/compiletest/ui/attention_kernel.rs`	—	PASS
`swiglu`	`tests/compiletest/ui/attention_kernel.rs`	`swiglu.py`	PASS
`geglu`	`tests/compiletest/ui/attention_kernel.rs`	—	PASS
`masked_fill`	`tests/compiletest/ui/attention_kernel.rs`	`masked_fill.py`	PASS
`causal_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	—	PASS
`cross_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`cross_attention.py`	PASS
`multi_query_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`multi_query_attention.py`	PASS
`group_query_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`group_query_attention.py`	PASS
`kv_cached_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	—	PASS
`cross_modal_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`cross_modal_attention.py`	PASS
`linear_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`linear_attention.py`	PASS
`sparse_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`sparse_attention.py`	PASS
`windowed_causal_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`windowed_causal_attention.py`	PASS
`min_gpt_causal_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`min_gpt_causal_attention.py`	PASS
`relu_self_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`relu_self_attention.py`	PASS
`vision_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`vision_attention.py`	PASS
`scaled_dot_product_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`scaled_dot_product_attention.py`	PASS
`sdpa_inference`	`tests/compiletest/ui/attention_extended_kernel.rs`	`sdpa_inference.py`	PASS
`sdpa_long_context`	`tests/compiletest/ui/attention_extended_kernel.rs`	`sdpa_long_context.py`	PASS
`kv_cached_chat_batch_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`kv_cached_chat_batch_attention.py`	PASS
`kv_cached_speculative_attention`	`tests/compiletest/ui/attention_extended_kernel.rs`	`kv_cached_speculative_attention.py`	PASS

Broadcast (12 kernels)

Applicable vulnerability patterns: V1(type erasure),V2(bounds),V5(double free)

MKB reference: reference_kernels/broadcast/

Kernel Function	Source File	MKB Reference	910B3 Status
`add_bias`	`tests/compiletest/ui/broadcast_ops_kernel.rs`	`add_bias.py`	PASS
`elementwise_mul`	`tests/compiletest/ui/broadcast_ops_kernel.rs`	`elementwise_mul.py`	PASS
`elementwise_div`	`tests/compiletest/ui/broadcast_ops_kernel.rs`	`elementwise_div.py`	PASS
`elementwise_sub`	`tests/compiletest/ui/broadcast_ops_kernel.rs`	`elementwise_sub.py`	PASS
`elementwise_max`	`tests/compiletest/ui/broadcast_ops_kernel.rs`	`elementwise_max.py`	PASS
`clamp`	`tests/compiletest/ui/broadcast_ops_kernel.rs`	—	PASS
`elementwise_min`	`tests/compiletest/ui/broadcast_ops_kernel.rs`	`elementwise_min.py`	PASS
`elementwise_square`	`tests/compiletest/ui/broadcast_ops_kernel.rs`	—	PASS
`where_broadcast`	`tests/compiletest/ui/broadcast_ext_kernel.rs`	—	PASS
`logic_and_broadcast`	`tests/compiletest/ui/broadcast_ext_kernel.rs`	`logic_and_broadcast.py`	PASS
`power_broadcast`	`tests/compiletest/ui/broadcast_ext_kernel.rs`	`power_broadcast.py`	PASS
`scalar_mul`	`tests/compiletest/ui/scalar_mul_kernel.rs`	`scalar_mul.py`	PASS

Convolution (34 kernels)

Applicable vulnerability patterns: V2(nested loop OOB),V3(stride*index overflow)

MKB reference: reference_kernels/convolution/

Kernel Function	Source File	MKB Reference	910B3 Status
`conv_standard_1d`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_1d.py`	PASS
`conv_standard_1d_dilated_strided`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_1d_dilated_strided.py`	PASS
`conv_standard_2d_square_square`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_2d_square_square.py`	PASS
`conv_standard_2d_asym_square`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_2d_asym_square.py`	PASS
`conv_standard_2d_square_asym`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_2d_square_asym.py`	PASS
`conv_standard_2d_asym_asym`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_2d_asym_asym.py`	PASS
`conv_standard_2d_dilated_padded`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_2d_dilated_padded.py`	PASS
`conv_standard_3d_square_square`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_3d_square_square.py`	PASS
`conv_standard_3d_asym_square`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_3d_asym_square.py`	PASS
`conv_standard_3d_square_asym`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_3d_square_asym.py`	PASS
`conv_standard_3d_asym_asym`	`tests/compiletest/ui/conv_standard_kernel.rs`	`conv_standard_3d_asym_asym.py`	PASS
`conv_depthwise_2d_sq_sq`	`tests/compiletest/ui/conv_depthwise_kernel.rs`	`conv_depthwise_2d_sq_sq.py`	PASS
`conv_depthwise_2d_asym_sq`	`tests/compiletest/ui/conv_depthwise_kernel.rs`	`conv_depthwise_2d_asym_sq.py`	PASS
`conv_depthwise_2d_sq_asym`	`tests/compiletest/ui/conv_depthwise_kernel.rs`	`conv_depthwise_2d_sq_asym.py`	PASS
`conv_depthwise_2d_asym_asym`	`tests/compiletest/ui/conv_depthwise_kernel.rs`	`conv_depthwise_2d_asym_asym.py`	PASS
`conv_depthwise_separable_2d`	`tests/compiletest/ui/conv_depthwise_kernel.rs`	`conv_depthwise_separable_2d.py`	PASS
`conv_pointwise_2d`	`tests/compiletest/ui/conv_depthwise_kernel.rs`	`conv_pointwise_2d.py`	PASS
`conv_transposed_1d`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_1d.py`	PASS
`conv_transposed_1d_dilated`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_1d_dilated.py`	PASS
`conv_transposed_1d_asym_padded_strided_dilated`	`tests/compiletest/ui/conv_transpose_kernel.rs`	—	PASS
`conv_transposed_2d_sq_sq`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_2d_sq_sq.py`	PASS
`conv_transposed_2d_sq_asym`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_2d_sq_asym.py`	PASS
`conv_transposed_2d_asym_sq`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_2d_asym_sq.py`	PASS
`conv_transposed_2d_asym_asym`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_2d_asym_asym.py`	PASS
`conv_transposed_2d_asym_asym_padded`	`tests/compiletest/ui/conv_transpose_kernel.rs`	—	PASS
`conv_transposed_2d_dilated_padded_strided`	`tests/compiletest/ui/conv_transpose_kernel.rs`	—	PASS
`conv_transposed_2d_grouped`	`tests/compiletest/ui/conv_transpose_kernel.rs`	—	PASS
`conv_transposed_3d_sq_sq`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_3d_sq_sq.py`	PASS
`conv_transposed_3d_sq_asym`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_3d_sq_asym.py`	PASS
`conv_transposed_3d_asym_sq`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_3d_asym_sq.py`	PASS
`conv_transposed_3d_asym_asym`	`tests/compiletest/ui/conv_transpose_kernel.rs`	`conv_transposed_3d_asym_asym.py`	PASS
`conv_transposed_3d_asym_sq_grouped`	`tests/compiletest/ui/conv_transpose_kernel.rs`	—	PASS
`conv_transposed_3d_asym_asym_grouped`	`tests/compiletest/ui/conv_transpose_kernel.rs`	—	PASS
`conv_transposed_3d_sq_sq_dilated`	`tests/compiletest/ui/conv_transpose_kernel.rs`	—	PASS

Fuse (120 kernels)

Applicable vulnerability patterns: V1,V2,V4(use-after-free in chain),V6(inter-op sync)

MKB reference: reference_kernels/fuse/

Kernel Function	Source File	MKB Reference	910B3 Status
`fused_relu_hardswish`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_relu_hardswish.py`	PASS
`fused_hardswish_relu`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_hardswish_relu.py`	PASS
`fused_mish_mish`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_mish_mish.py`	PASS
`fused_mish_tanh`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_mish_tanh.py`	PASS
`fused_min_tanh_tanh`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_min_tanh_tanh.py`	PASS
`fused_mul_leakyrelu_gelu`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_mul_leakyrelu_gelu.py`	PASS
`fused_sub_tanh_sub`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_sub_tanh_sub.py`	PASS
`fused_sigmoid_sum`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_sigmoid_sum.py`	PASS
`fused_add_scale_sigmoid`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_add_scale_sigmoid.py`	PASS
`fused_scale_min`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_scale_min.py`	PASS
`fused_leakyrelu_leakyrelu_gelu_gelu`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_leakyrelu_leakyrelu_gelu_gelu.py`	PASS
`fused_divide_leakyrelu`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_divide_leakyrelu.py`	PASS
`fused_sub_hardswish`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_sub_hardswish.py`	PASS
`fused_tanh_scale_bias_max`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_tanh_scale_bias_max.py`	PASS
`fused_relu_bias_add`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_relu_bias_add.py`	PASS
`fused_hardswish_relu_softmax_mean`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_hardswish_relu_softmax_mean.py`	PASS
`fused_leakyrelu_clamp_gelu`	`tests/compiletest/ui/fused_activation_chain_kernel.rs`	`fused_leakyrelu_clamp_gelu.py`	PASS
`fused_norm_add_mul`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_norm_add_mul.py`	PASS
`fused_scale_norm`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_scale_norm.py`	PASS
`fused_sub_mish_mish`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_sub_mish_mish.py`	PASS
`fused_sub_tanh_sub_mean`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_sub_tanh_sub_mean.py`	PASS
`fused_min_add_mul`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_min_add_mul.py`	PASS
`fused_elu_scale`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_elu_scale.py`	PASS
`fused_selu_add`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_selu_add.py`	PASS
`fused_softplus_tanh`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_softplus_tanh.py`	PASS
`fused_relu_scale_add`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_relu_scale_add.py`	PASS
`fused_sigmoid_gate`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_sigmoid_gate.py`	PASS
`fused_exp_reduce_sum`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_exp_reduce_sum.py`	PASS
`log_sum_exp`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`log_sum_exp.py`	PASS
`fused_max_lse_relu`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_max_lse_relu.py`	PASS
`fused_hardswish_gelu`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_hardswish_gelu.py`	PASS
`fused_softsign_scale_add`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_softsign_scale_add.py`	PASS
`fused_hardsigmoid_scale_clamp`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_hardsigmoid_scale_clamp.py`	PASS
`fused_abs_sum`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_abs_sum.py`	PASS
`fused_rmsnorm_mish_scale`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_rmsnorm_mish_scale.py`	PASS
`fused_reciprocal_scale_add`	`tests/compiletest/ui/fused_multi_op_kernel.rs`	`fused_reciprocal_scale_add.py`	PASS
`fused_layernorm_relu`	`tests/compiletest/ui/fused_norm_activation_kernel.rs`	`fused_layernorm_relu.py`	PASS
`fused_layernorm_sigmoid`	`tests/compiletest/ui/fused_norm_activation_kernel.rs`	`fused_layernorm_sigmoid.py`	PASS
`fused_rmsnorm_swish`	`tests/compiletest/ui/fused_norm_activation_kernel.rs`	`fused_rmsnorm_swish.py`	PASS
`fused_layernorm_tanh_hardswish`	`tests/compiletest/ui/fused_norm_activation_kernel.rs`	`fused_layernorm_tanh_hardswish.py`	PASS
`fused_softmax_mean`	`tests/compiletest/ui/fused_norm_activation_kernel.rs`	`fused_softmax_mean.py`	PASS
`fused_layernorm_gelu`	`tests/compiletest/ui/fused_norm_activation_kernel.rs`	`fused_layernorm_gelu.py`	PASS
`fused_rmsnorm_gelu`	`tests/compiletest/ui/fused_norm_activation_kernel.rs`	`fused_rmsnorm_gelu.py`	PASS
`fused_log_softmax_mean`	`tests/compiletest/ui/fused_norm_activation_kernel.rs`	`fused_log_softmax_mean.py`	PASS
`test_sigmoid`	`tests/compiletest/ui/composite_ops_kernel.rs`	—	PASS
`test_tanh`	`tests/compiletest/ui/composite_ops_kernel.rs`	—	PASS
`test_gelu`	`tests/compiletest/ui/composite_ops_kernel.rs`	—	PASS
`test_softmax`	`tests/compiletest/ui/composite_ops_kernel.rs`	—	PASS
`conv2d_activation_batch_norm`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_activation_batch_norm.py`	PASS
`conv2d_add_scale_sigmoid_group_norm`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_add_scale_sigmoid_group_norm.py`	PASS
`conv2d_avg_pool_sigmoid_sum`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_avg_pool_sigmoid_sum.py`	PASS
`conv2d_batch_norm_scaling`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_batch_norm_scaling.py`	PASS
`conv2d_gelu_global_avg_pool`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_gelu_global_avg_pool.py`	PASS
`conv2d_group_norm_scale_max_pool_clamp`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_group_norm_scale_max_pool_clamp.py`	PASS
`conv2d_group_norm_tanh_hard_swish_residual_add_log_sum_exp`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_group_norm_tanh_hard_swish_residual_add_log_sum_exp.py`	PASS
`conv2d_instance_norm_divide`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_instance_norm_divide.py`	PASS
`conv2d_subtract_hard_swish_max_pool_mish`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_subtract_hard_swish_max_pool_mish.py`	PASS
`conv2d_subtract_subtract_mish`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_subtract_subtract_mish.py`	PASS
`conv2d_subtract_tanh_subtract_avg_pool`	`tests/compiletest/ui/fused_conv2d_ext_kernel.rs`	`conv2d_subtract_tanh_subtract_avg_pool.py`	PASS
`conv3d_divide_max_global_avg_pool_bias_add_sum`	`tests/compiletest/ui/fused_conv3d_ext_kernel.rs`	`conv3d_divide_max_global_avg_pool_bias_add_sum.py`	PASS
`conv3d_leaky_relu_sum_clamp_gelu`	`tests/compiletest/ui/fused_conv3d_ext_kernel.rs`	`conv3d_leaky_relu_sum_clamp_gelu.py`	PASS
`conv3d_multiply_instance_norm_clamp_multiply_max`	`tests/compiletest/ui/fused_conv3d_ext_kernel.rs`	`conv3d_multiply_instance_norm_clamp_multiply_max.py`	PASS
`conv3d_relu_leaky_relu_gelu_sigmoid_bias_add`	`tests/compiletest/ui/fused_conv3d_ext_kernel.rs`	`conv3d_relu_leaky_relu_gelu_sigmoid_bias_add.py`	PASS
`conv3d_scaling_tanh_multiply_sigmoid`	`tests/compiletest/ui/fused_conv3d_ext_kernel.rs`	`conv3d_scaling_tanh_multiply_sigmoid.py`	PASS
`conv3d_softmax_max_pool_max_pool`	`tests/compiletest/ui/fused_conv3d_ext_kernel.rs`	`conv3d_softmax_max_pool_max_pool.py`	PASS
`conv_transpose2d_add_min_gelu_multiply`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`conv_transpose2d_add_min_gelu_multiply.py`	PASS
`conv_transpose2d_bias_add_clamp_scaling_clamp_divide`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`conv_transpose2d_bias_add_clamp_scaling_clamp_divide.py`	PASS
`conv_transpose2d_gelu_group_norm`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`conv_transpose2d_gelu_group_norm.py`	PASS
`conv_transpose2d_max_pool_hardtanh_mean_tanh`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`conv_transpose2d_max_pool_hardtanh_mean_tanh.py`	PASS
`conv_transpose2d_min_sum_gelu_add`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`conv_transpose2d_min_sum_gelu_add.py`	PASS
`conv_transpose2d_mish_add_hardtanh_scaling`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`conv_transpose2d_mish_add_hardtanh_scaling.py`	PASS
`conv_transpose2d_multiply_global_avg_pool_global_avg_pool_mean`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`conv_transpose2d_multiply_global_avg_pool_global_avg_pool_mean.py`	PASS
`conv_transpose2d_subtract_tanh`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`conv_transpose2d_subtract_tanh.py`	PASS
`convtranspose2d_batchnorm_tanh_maxpool_groupnorm`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`convtranspose2d_batchnorm_tanh_maxpool_groupnorm.py`	PASS
`convtranspose2d_globalavgpool_biasadd_logsumexp_sum_multiply`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`convtranspose2d_globalavgpool_biasadd_logsumexp_sum_multiply.py`	PASS
`convtranspose2d_softmax_biasadd_scaling_sigmoid`	`tests/compiletest/ui/fused_conv_transpose2d_kernel.rs`	`convtranspose2d_softmax_biasadd_scaling_sigmoid.py`	PASS
`conv_transpose3d_add_hard_swish`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_add_hard_swish.py`	PASS
`conv_transpose3d_avg_pool_clamp_softmax_multiply`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_avg_pool_clamp_softmax_multiply.py`	PASS
`conv_transpose3d_batch_norm_avg_pool_avg_pool`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_batch_norm_avg_pool_avg_pool.py`	PASS
`conv_transpose3d_batch_norm_subtract`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_batch_norm_subtract.py`	PASS
`conv_transpose3d_clamp_min_divide`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_clamp_min_divide.py`	PASS
`conv_transpose3d_layer_norm_gelu_scaling`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_layer_norm_gelu_scaling.py`	PASS
`conv_transpose3d_leaky_relu_multiply_leaky_relu_max`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_leaky_relu_multiply_leaky_relu_max.py`	PASS
`conv_transpose3d_log_sum_exp_hard_swish_subtract_clamp_max`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_log_sum_exp_hard_swish_subtract_clamp_max.py`	PASS
`conv_transpose3d_max_max_sum`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_max_max_sum.py`	PASS
`conv_transpose3d_max_pool_softmax_subtract_swish_max`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_max_pool_softmax_subtract_swish_max.py`	PASS
`conv_transpose3d_multiply_max_global_avg_pool_clamp`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_multiply_max_global_avg_pool_clamp.py`	PASS
`conv_transpose3d_scale_batch_norm_global_avg_pool`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_scale_batch_norm_global_avg_pool.py`	PASS
`conv_transpose3d_scaling_avg_pool_bias_add_scaling`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_scaling_avg_pool_bias_add_scaling.py`	PASS
`conv_transpose3d_softmax_sigmoid`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_softmax_sigmoid.py`	PASS
`conv_transpose3d_sum_layer_norm_avg_pool_gelu`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_sum_layer_norm_avg_pool_gelu.py`	PASS
`conv_transpose3d_sum_residual_add_multiply_residual_add`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_sum_residual_add_multiply_residual_add.py`	PASS
`conv_transpose3d_swish_group_norm_hard_swish`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`conv_transpose3d_swish_group_norm_hard_swish.py`	PASS
`convtranspose3d_mean_add_softmax_tanh_scaling`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`convtranspose3d_mean_add_softmax_tanh_scaling.py`	PASS
`convtranspose3d_relu_groupnorm`	`tests/compiletest/ui/fused_conv_transpose3d_kernel.rs`	`convtranspose3d_relu_groupnorm.py`	PASS
`gemm_add_relu`	`tests/compiletest/ui/fused_gemm_ext_kernel.rs`	`gemm_add_relu.py`	PASS
`gemm_batch_norm_gelu_group_norm_mean_relu`	`tests/compiletest/ui/fused_gemm_ext_kernel.rs`	`gemm_batch_norm_gelu_group_norm_mean_relu.py`	PASS
`gemm_batch_norm_scaling_softmax`	`tests/compiletest/ui/fused_gemm_ext_kernel.rs`	`gemm_batch_norm_scaling_softmax.py`	PASS
`gemm_log_sum_exp_leaky_relu_leaky_relu_gelu_gelu`	`tests/compiletest/ui/fused_gemm_ext_kernel.rs`	`gemm_log_sum_exp_leaky_relu_leaky_relu_gelu_gelu.py`	PASS
`gemm_sigmoid_sum_log_sum_exp`	`tests/compiletest/ui/fused_gemm_ext_kernel.rs`	`gemm_sigmoid_sum_log_sum_exp.py`	PASS
`gemm_subtract_global_avg_pool_log_sum_exp_gelu_residual_add`	`tests/compiletest/ui/fused_gemm_ext_kernel.rs`	`gemm_subtract_global_avg_pool_log_sum_exp_gelu_residual_add.py`	PASS
`matmul_avg_pool_gelu_scale_max`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_avg_pool_gelu_scale_max.py`	PASS
`matmul_batch_norm_bias_add_divide_swish`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_batch_norm_bias_add_divide_swish.py`	PASS
`matmul_dropout_mean_softmax`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_dropout_mean_softmax.py`	PASS
`matmul_scale_residual_add_clamp_log_sum_exp_mish`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_scale_residual_add_clamp_log_sum_exp_mish.py`	PASS
`matmul_scaling_residual_add`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_scaling_residual_add.py`	PASS
`matmul_sigmoid_sum`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_sigmoid_sum.py`	PASS
`matmul_subtract_multiply_relu`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_subtract_multiply_relu.py`	PASS
`matmul_sum_max_avg_pool_log_sum_exp_log_sum_exp`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_sum_max_avg_pool_log_sum_exp_log_sum_exp.py`	PASS
`matmul_swish_scaling`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_swish_scaling.py`	PASS
`matmul_swish_sum_group_norm`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`matmul_swish_sum_group_norm.py`	PASS
`bmm_instance_norm_sum_residual_add_multiply`	`tests/compiletest/ui/fused_matmul_ext_kernel.rs`	`bmm_instance_norm_sum_residual_add_multiply.py`	PASS
`fused_gemm_norm_gelu`	`tests/compiletest/ui/fused_matmul_norm_kernel.rs`	—	PASS
`fused_gemm_norm_scale_softmax`	`tests/compiletest/ui/fused_matmul_norm_kernel.rs`	—	PASS
`fused_gemm_scale_norm`	`tests/compiletest/ui/fused_matmul_norm_kernel.rs`	—	PASS
`fused_gemm_norm_hardtanh`	`tests/compiletest/ui/fused_matmul_norm_kernel.rs`	—	PASS
`fused_gemm_norm_swish_mul_swish`	`tests/compiletest/ui/fused_matmul_norm_kernel.rs`	—	PASS
`fused_gemm_bias_hardtanh_mish_norm`	`tests/compiletest/ui/fused_matmul_norm_kernel.rs`	—	PASS
`gemm_scale_batch_norm`	`tests/compiletest/ui/fused_matmul_norm_kernel.rs`	`gemm_scale_batch_norm.py`	PASS
`gemm_scale_batchnorm`	`tests/compiletest/ui/fused_matmul_norm_kernel.rs`	`gemm_scale_batchnorm.py`	PASS

Index (12 kernels)

Applicable vulnerability patterns: V2(gather/scatter OOB),V3(index calc overflow)

MKB reference: reference_kernels/index/

Kernel Function	Source File	MKB Reference	910B3 Status
`argmax`	`tests/compiletest/ui/index_ops_kernel.rs`	`argmax.py`	PASS
`argmin`	`tests/compiletest/ui/index_ops_kernel.rs`	`argmin.py`	PASS
`gather`	`tests/compiletest/ui/index_ops_kernel.rs`	`gather.py`	PASS
`scatter`	`tests/compiletest/ui/index_ops_kernel.rs`	`scatter.py`	PASS
`scatter_add`	`tests/compiletest/ui/index_ops_kernel.rs`	`scatter_add.py`	PASS
`index_select`	`tests/compiletest/ui/index_ops_kernel.rs`	`index_select.py`	PASS
`index_copy`	`tests/compiletest/ui/index_ops_kernel.rs`	`index_copy.py`	PASS
`index_add`	`tests/compiletest/ui/index_ops_kernel.rs`	`index_add.py`	PASS
`embedding`	`tests/compiletest/ui/index_ops_kernel.rs`	`embedding.py`	PASS
`masked_fill`	`tests/compiletest/ui/index_ops_kernel.rs`	`masked_fill.py`	PASS
`inplace_update`	`tests/compiletest/ui/index_ops_kernel.rs`	`inplace_update.py`	PASS
`take_along_dim`	`tests/compiletest/ui/index_ops_kernel.rs`	`take_along_dim.py`	PASS

Loss (6 kernels)

Applicable vulnerability patterns: V1,V2,V6(reduction sync)

MKB reference: reference_kernels/loss/

Kernel Function	Source File	MKB Reference	910B3 Status
`mse_loss`	`tests/compiletest/ui/loss_ops_kernel.rs`	`mse_loss.py`	PASS
`huber_loss`	`tests/compiletest/ui/loss_ops_kernel.rs`	`huber_loss.py`	PASS
`hinge_loss`	`tests/compiletest/ui/loss_ops_kernel.rs`	`hinge_loss.py`	PASS
`cosine_similarity`	`tests/compiletest/ui/loss_ops_kernel.rs`	`cosine_similarity.py`	PASS
`cross_entropy_loss`	`tests/compiletest/ui/loss_ops_kernel.rs`	`cross_entropy_loss.py`	PASS
`kl_div_loss`	`tests/compiletest/ui/loss_ops_kernel.rs`	`kl_div_loss.py`	PASS

Math (5 kernels)

Applicable vulnerability patterns: V2(cumulative bounds),V3(offset overflow)

MKB reference: reference_kernels/math/

Kernel Function	Source File	MKB Reference	910B3 Status
`matrix_scalar_mul`	`tests/compiletest/ui/math_ops_kernel.rs`	`matrix_scalar_mul.py`	PASS
`cumprod`	`tests/compiletest/ui/math_cumulative_kernel.rs`	`cumprod.py`	PASS
`cumsum`	`tests/compiletest/ui/math_cumulative_kernel.rs`	`cumsum.py`	PASS
`cumsum_exclusive`	`tests/compiletest/ui/math_cumulative_kernel.rs`	`cumsum_exclusive.py`	PASS
`cumsum_reverse`	`tests/compiletest/ui/math_cumulative_kernel.rs`	`cumsum_reverse.py`	PASS

Matmul (23 kernels)

Applicable vulnerability patterns: V1(type erasure f16/f32),V2(tile bounds),V3(dim overflow),V6(cube sync)

MKB reference: reference_kernels/matmul/

Kernel Function	Source File	MKB Reference	910B3 Status
`matmul`	`tests/compiletest/ui/matmul_kernel.rs`	`matmul.py`	PASS
`matmul_standard`	`tests/compiletest/ui/matmul_ops_kernel.rs`	`matmul_standard.py`	PASS
`matmul_square`	`tests/compiletest/ui/matmul_ops_kernel.rs`	`matmul_square.py`	PASS
`matmul_matvec`	`tests/compiletest/ui/matmul_ops_kernel.rs`	`matmul_matvec.py`	PASS
`matmul_large_k`	`tests/compiletest/ui/matmul_ops_kernel.rs`	`matmul_large_k.py`	PASS
`matmul_small_k`	`tests/compiletest/ui/matmul_ops_kernel.rs`	`matmul_small_k.py`	PASS
`matmul_irregular`	`tests/compiletest/ui/matmul_ops_kernel.rs`	`matmul_irregular.py`	PASS
`matmul_tall_skinny`	`tests/compiletest/ui/matmul_ops_kernel.rs`	`matmul_tall_skinny.py`	PASS
`matmul_transposed_a`	`tests/compiletest/ui/matmul_transpose_kernel.rs`	—	PASS
`matmul_transposed_b`	`tests/compiletest/ui/matmul_transpose_kernel.rs`	—	PASS
`matmul_transposed_both`	`tests/compiletest/ui/matmul_transpose_kernel.rs`	—	PASS
`matmul_lower_triangular`	`tests/compiletest/ui/matmul_transpose_kernel.rs`	—	PASS
`matmul_upper_triangular`	`tests/compiletest/ui/matmul_transpose_kernel.rs`	—	PASS
`matmul_batched`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`matmul_batched.py`	PASS
`matmul_symmetric`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`matmul_symmetric.py`	PASS
`matmul_bias`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`matmul_bias.py`	PASS
`matmul_scaled`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`matmul_scaled.py`	PASS
`gemm_full`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`gemm_full.py`	PASS
`matmul_wide`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`matmul_wide.py`	PASS
`matmul_relu_matmul`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`matmul_relu_matmul.py`	PASS
`matmul_accumulate`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`matmul_accumulate.py`	PASS
`matmul_diag_scale`	`tests/compiletest/ui/matmul_extended_kernel.rs`	`matmul_diag_scale.py`	PASS
`outer_product`	`tests/compiletest/ui/matmul_extended_kernel.rs`	—	PASS

Normalization (10 kernels)

Applicable vulnerability patterns: V1,V2,V6(reduce-normalize sync)

MKB reference: reference_kernels/normalization/

Kernel Function	Source File	MKB Reference	910B3 Status
`rms_norm`	`tests/compiletest/ui/norm_ops_kernel.rs`	`rms_norm.py`	PASS
`l1_norm`	`tests/compiletest/ui/norm_ops_kernel.rs`	`l1_norm.py`	PASS
`l2_norm`	`tests/compiletest/ui/norm_ops_kernel.rs`	`l2_norm.py`	PASS
`l2_normalize`	`tests/compiletest/ui/norm_ops_kernel.rs`	`l2_normalize.py`	PASS
`layer_norm`	`tests/compiletest/ui/norm_ops_kernel.rs`	`layer_norm.py`	PASS
`batch_norm`	`tests/compiletest/ui/norm_extended_kernel.rs`	—	PASS
`group_norm`	`tests/compiletest/ui/norm_extended_kernel.rs`	`group_norm.py`	PASS
`instance_norm`	`tests/compiletest/ui/norm_extended_kernel.rs`	`instance_norm.py`	PASS
`frobenius_norm`	`tests/compiletest/ui/norm_extended_kernel.rs`	`frobenius_norm.py`	PASS
`layernorm`	`tests/compiletest/ui/layernorm_kernel.rs`	`layernorm.py`	PASS

Optimizer (6 kernels)

Applicable vulnerability patterns: V1,V2(param bounds),V4(in-place update UAF)

MKB reference: reference_kernels/optimizer/

Kernel Function	Source File	MKB Reference	910B3 Status
`sgd_update`	`tests/compiletest/ui/optimizer_ops_kernel.rs`	`sgd_update.py`	PASS
`sgd_momentum`	`tests/compiletest/ui/optimizer_ops_kernel.rs`	`sgd_momentum.py`	PASS
`adagrad_update`	`tests/compiletest/ui/optimizer_ops_kernel.rs`	`adagrad_update.py`	PASS
`rmsprop_update`	`tests/compiletest/ui/optimizer_ops_kernel.rs`	`rmsprop_update.py`	PASS
`adam_update`	`tests/compiletest/ui/optimizer_ops_kernel.rs`	`adam_update.py`	PASS
`lamb_update`	`tests/compiletest/ui/optimizer_ext_kernel.rs`	`lamb_update.py`	PASS

Pooling (12 kernels)

Applicable vulnerability patterns: V2(window OOB),V3(stride overflow)

MKB reference: reference_kernels/pooling/

Kernel Function	Source File	MKB Reference	910B3 Status
`global_avg_pool`	`tests/compiletest/ui/pooling_ops_kernel.rs`	`global_avg_pool.py`	PASS
`global_max_pool`	`tests/compiletest/ui/pooling_ops_kernel.rs`	`global_max_pool.py`	PASS
`global_min_pool`	`tests/compiletest/ui/pooling_ops_kernel.rs`	`global_min_pool.py`	PASS
`fused_avgpool_sigmoid`	`tests/compiletest/ui/pooling_ops_kernel.rs`	`fused_avgpool_sigmoid.py`	PASS
`fused_pool_sigmoid_sum`	`tests/compiletest/ui/pooling_ops_kernel.rs`	`fused_pool_sigmoid_sum.py`	PASS
`lp_pool_2`	`tests/compiletest/ui/pooling_ops_kernel.rs`	`lp_pool_2.py`	PASS
`max_pooling_1d`	`tests/compiletest/ui/pooling_windowed_kernel.rs`	`max_pooling_1d.py`	PASS
`max_pooling_2d`	`tests/compiletest/ui/pooling_windowed_kernel.rs`	`max_pooling_2d.py`	PASS
`max_pooling_3d`	`tests/compiletest/ui/pooling_windowed_kernel.rs`	`max_pooling_3d.py`	PASS
`average_pooling_1d`	`tests/compiletest/ui/pooling_windowed_kernel.rs`	`average_pooling_1d.py`	PASS
`average_pooling_2d`	`tests/compiletest/ui/pooling_windowed_kernel.rs`	`average_pooling_2d.py`	PASS
`average_pooling_3d`	`tests/compiletest/ui/pooling_windowed_kernel.rs`	`average_pooling_3d.py`	PASS

Reduce (5 kernels)

Applicable vulnerability patterns: V1,V2,V6(reduction pipeline sync)

MKB reference: reference_kernels/reduce/

Kernel Function	Source File	MKB Reference	910B3 Status
`reduce_max`	`tests/compiletest/ui/reduce_ops_kernel.rs`	`reduce_max.py`	PASS
`reduce_min`	`tests/compiletest/ui/reduce_ops_kernel.rs`	`reduce_min.py`	PASS
`reduce_sum`	`tests/compiletest/ui/reduce_ops_kernel.rs`	`reduce_sum.py`	PASS
`reduce_mean`	`tests/compiletest/ui/reduce_ops_kernel.rs`	`reduce_mean.py`	PASS
`reduce_prod`	`tests/compiletest/ui/reduce_ops_kernel.rs`	`reduce_prod.py`	PASS

Resize (15 kernels)

Applicable vulnerability patterns: V2(interpolation OOB),V3(coordinate overflow)

MKB reference: reference_kernels/resize/

Kernel Function	Source File	MKB Reference	910B3 Status
`resize_nearest`	`tests/compiletest/ui/resize_ops_kernel.rs`	`resize_nearest.py`	PASS
`lerp`	`tests/compiletest/ui/resize_ops_kernel.rs`	—	PASS
`bicubic_weight`	`tests/compiletest/ui/resize_ops_kernel.rs`	`bicubic_weight.py`	PASS
`weighted_sum`	`tests/compiletest/ui/resize_ops_kernel.rs`	—	PASS
`trilinear_1d`	`tests/compiletest/ui/resize_ops_kernel.rs`	—	PASS
`bilinear_upsample_2d`	`tests/compiletest/ui/resize_spatial_kernel.rs`	`bilinear_upsample_2d.py`	PASS
`bicubic_upsample_2d`	`tests/compiletest/ui/resize_spatial_kernel.rs`	`bicubic_upsample_2d.py`	PASS
`nearest_upsample_2d`	`tests/compiletest/ui/resize_spatial_kernel.rs`	`nearest_upsample_2d.py`	PASS
`trilinear_upsample_3d`	`tests/compiletest/ui/resize_spatial_kernel.rs`	`trilinear_upsample_3d.py`	PASS
`downsample_bilinear_2d`	`tests/compiletest/ui/resize_spatial_kernel.rs`	`downsample_bilinear_2d.py`	PASS
`grid_sample_affine`	`tests/compiletest/ui/resize_ext_kernel.rs`	`grid_sample_affine.py`	PASS
`grid_sample_random_warp`	`tests/compiletest/ui/resize_ext_kernel.rs`	`grid_sample_random_warp.py`	PASS
`interpolate_dynamic`	`tests/compiletest/ui/resize_ext_kernel.rs`	`interpolate_dynamic.py`	PASS
`resize_with_antialias`	`tests/compiletest/ui/resize_ext_kernel.rs`	`resize_with_antialias.py`	PASS
`upsample_grid_sample`	`tests/compiletest/ui/resize_ext_kernel.rs`	`upsample_grid_sample.py`	PASS

Tiled (16 kernels)

Applicable vulnerability patterns: V2(tile boundary OOB),V6(tile-boundary sync)

Kernel Function	Source File	910B3 Status
`relu_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`sigmoid_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`gelu_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`tanh_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`swish_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`exp_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`vec_add_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`vec_mul_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`elu_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`mish_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`layernorm_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`softmax_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`selu_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`leaky_relu_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`hardswish_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS
`rmsnorm_tiled`	`tests/compiletest/ui/tiled_kernel.rs`	PASS

Multiblock (16 kernels)

Applicable vulnerability patterns: V2(block partition OOB),V6(cross-block sync)

Kernel Function	Source File	910B3 Status
`relu_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`sigmoid_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`gelu_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`tanh_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`softmax_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`layernorm_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`vec_add_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`mish_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`swish_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`elu_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`selu_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`leaky_relu_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`rmsnorm_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`hardswish_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`hardsigmoid_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS
`softplus_multiblock`	`tests/compiletest/ui/multiblock_kernel.rs`	PASS

F16 (14 kernels)

Applicable vulnerability patterns: V1(f16/f32 type confusion)

Kernel Function	Source File	910B3 Status
`relu_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`sigmoid_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`abs_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`exp_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`ln_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`sqrt_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`rsqrt_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`reciprocal_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`vec_add_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`vec_sub_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`vec_mul_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`vec_div_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`reduce_max_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS
`reduce_sum_f16`	`tests/compiletest/ui/f16_activation_kernel.rs`	PASS

Unary_math (8 kernels)

Applicable vulnerability patterns: V1,V2

Kernel Function	Source File	910B3 Status
`exp_f32`	`tests/compiletest/ui/f32_unary_kernel.rs`	PASS
`ln_f32`	`tests/compiletest/ui/f32_unary_kernel.rs`	PASS
`sqrt_f32`	`tests/compiletest/ui/f32_unary_kernel.rs`	PASS
`rsqrt_f32`	`tests/compiletest/ui/f32_unary_kernel.rs`	PASS
`reciprocal_f32`	`tests/compiletest/ui/f32_unary_kernel.rs`	PASS
`negate_f32`	`tests/compiletest/ui/f32_unary_kernel.rs`	PASS
`square_f32`	`tests/compiletest/ui/f32_unary_kernel.rs`	PASS
`cube_f32`	`tests/compiletest/ui/f32_unary_kernel.rs`	PASS

Deployable Kernels (with host code)

Kernel	Source File	Purpose
`?`	`examples/bench_softmax_tile/kernels/src/lib.rs`	Deployable kernel
`scale_f16`	`examples/mha/kernels/src/lib.rs`	Multi-head attention (f16 scale + softmax)
`softmax_rows_f16`	`examples/mha/kernels/src/lib.rs`	Multi-head attention (f16 scale + softmax)
`?`	`examples/acl_vec_mul/kernels/src/lib.rs`	Vector multiplication example
`mul`	`examples/acl_vec_mul/kernels/src/lib.rs`	Vector multiplication example
`tile_softmax`	`examples/tile_softmax/kernels/src/lib.rs`	Deployable kernel
`?`	`examples/tile_matmul/kernels/src/lib.rs`	Deployable kernel
`softmax`	`examples/bench_softmax_rs/kernels/src/lib.rs`	Softmax benchmark (Rust)
`add`	`examples/acl_rs_vec_add/kernels/src/lib.rs`	Vector addition end-to-end example
`test_store_const`	`examples/acl_softmax/kernels/src/lib.rs`	Softmax with store/copy test kernels
`test_copy`	`examples/acl_softmax/kernels/src/lib.rs`	Softmax with store/copy test kernels
`softmax`	`examples/acl_softmax/kernels/src/lib.rs`	Softmax with store/copy test kernels
`vec_add`	`examples/bench_vec_add_rs/kernels/src/lib.rs`	Vector add benchmark (Rust)
`matmul`	`examples/bench_matmul_rs/kernels/src/lib.rs`	Matrix multiply benchmark (Rust)

Memory Safety Case Studies

Each case pairs a vulnerable C++ kernel with a structurally safe Rust kernel.

Case	Vulnerability	C++ File	Rust File
1	Type confusion (`GM_ADDR` type erasure)	`vulnerable.cpp`	`safe.rs`
2	Buffer overflow (unchecked indexing)	`vulnerable.cpp`	`safe.rs`
3	Use-after-free (`FreeTensor` then access)	`vulnerable.cpp`	`safe.rs`
4	Missing sync (forgotten `pipe_barrier`)	`vulnerable.cpp`	`safe.rs`
5	Double free (repeated `FreeTensor`)	`vulnerable.cpp`	`safe.rs`
6	Integer overflow (silent offset wrap)	`vulnerable.cpp`	`safe.rs`

Performance Comparison (in progress)

Kernel	ascend-rs Time	AscendC C++ Time	Ratio	Notes
softmax (256)	0.077 ms	0.078 ms	0.99x	Zero overhead
softmax (16384)	0.087 ms	0.089 ms	0.98x	Zero overhead
relu	—	—	—	Pending
matmul	—	—	—	Pending
layernorm	—	—	—	Pending
conv2d	—	—	—	Pending

Performance benchmarking experiments are in progress. This table will be updated as results become available.

This appendix was auto-generated by bash scripts/generate_kernel_appendix.sh. Kernel counts: 486 compiletests + 19 deployable = 505 total.

Keyboard shortcuts

ascend-rs: Memory-Safe NPU Kernel Programming in Rust