API

heliaCORE keeps the CMSIS-NN C API shape where compatibility matters, then adds Ambiq-tuned kernels and integration paths for HELIA workloads on Apollo-class targets. Start with the operator families below, filter to the kernels you care about, then drill into exact generated signatures when you need implementation detail.

positioningWhat it isA CMSIS-NN-compatible kernel layer packaged for Ambiq firmware and HELIA runtime integration. compatibilityHow it differsInherited CMSIS-NN APIs stay recognizable while Ambiq coverage fills model-graph gaps around real deployments. integrationHow it fits HELIAUse heliaCORE through neuralSPOT-X or direct CMake/CMSIS-Pack flows depending on your firmware stack.

Browse By Kernel Family

computeConvolutionConv2D, depthwise, transpose convolution, wrappers, and buffer helpers. computeFully connectedDense layers, batch matmul paths, and scratch sizing helpers. graph glueElementwiseAdd, sub, mul, square difference, min/max, mean, and arithmetic glue. selectionReduction and comparisonArgmin/argmax, min/max reductions, comparisons, means, and vector sums. activationActivationReLU, LeakyReLU, PReLU, Hard-Swish, Logistic, Tanh, and clamp. layoutData movementPad, reshape, transpose, concatenate, gather, resize, and strided slice. tailPooling, softmax, quantizationClassifier tail APIs plus dtype conversion and requantization utilities. sequenceSequenceLSTM and SVDF functions for temporal workloads. indexTypes and supportStructs, enums, files, and low-level support helpers in the full generated index.

For file-by-file browsing, structs, enums, and the complete generated tree, open the generated C API reference.

Convolution Functions

66 functions
arm_convolve_1_x_n_s4 arm_convolve_1_x_n_s4_get_buffer_size arm_convolve_1_x_n_s8 arm_convolve_1_x_n_s8_get_buffer_size arm_convolve_1x1_out_s8 arm_convolve_1x1_s16_ns_np_nd arm_convolve_1x1_s4 arm_convolve_1x1_s4_fast arm_convolve_1x1_s4_fast_get_buffer_size arm_convolve_1x1_s8 arm_convolve_1x1_s8_fast arm_convolve_1x1_s8_fast_get_buffer_size arm_convolve_even_s4 arm_convolve_s16 arm_convolve_s16_fast_small_kernel arm_convolve_s16_get_buffer_size arm_convolve_s16_group_ch_mult_1 arm_convolve_s4 arm_convolve_s4_get_buffer_size arm_convolve_s8 arm_convolve_s8_get_buffer_size arm_convolve_s8_get_weights_sum_size arm_convolve_weight_sum arm_convolve_wrapper_s16 arm_convolve_wrapper_s16_get_buffer_size arm_convolve_wrapper_s16_get_buffer_size_dsp arm_convolve_wrapper_s16_get_buffer_size_mve arm_convolve_wrapper_s4 arm_convolve_wrapper_s4_get_buffer_size arm_convolve_wrapper_s4_get_buffer_size_dsp arm_convolve_wrapper_s4_get_buffer_size_mve arm_convolve_wrapper_s8 arm_convolve_wrapper_s8_get_buffer_size arm_convolve_wrapper_s8_get_buffer_size_dsp arm_convolve_wrapper_s8_get_buffer_size_mve arm_depthwise_conv_3x3_s8 arm_depthwise_conv_fast_s16 arm_depthwise_conv_fast_s16_get_buffer_size arm_depthwise_conv_s16 arm_depthwise_conv_s4 arm_depthwise_conv_s4_opt arm_depthwise_conv_s4_opt_get_buffer_size arm_depthwise_conv_s8 arm_depthwise_conv_s8_opt arm_depthwise_conv_s8_opt_get_buffer_size arm_depthwise_conv_s8_opt_get_buffer_size_dsp arm_depthwise_conv_s8_opt_get_buffer_size_mve arm_depthwise_conv_wrapper_s16 arm_depthwise_conv_wrapper_s16_get_buffer_size arm_depthwise_conv_wrapper_s16_get_buffer_size_dsp arm_depthwise_conv_wrapper_s16_get_buffer_size_mve arm_depthwise_conv_wrapper_s4 arm_depthwise_conv_wrapper_s4_get_buffer_size arm_depthwise_conv_wrapper_s4_get_buffer_size_dsp arm_depthwise_conv_wrapper_s4_get_buffer_size_mve arm_depthwise_conv_wrapper_s8 arm_depthwise_conv_wrapper_s8_get_buffer_size arm_depthwise_conv_wrapper_s8_get_buffer_size_dsp arm_depthwise_conv_wrapper_s8_get_buffer_size_mve arm_depthwise_convolve_weight_sum arm_transpose_conv_s8 arm_transpose_conv_s8_get_buffer_size arm_transpose_conv_s8_get_buffer_size_dsp arm_transpose_conv_s8_get_buffer_size_mve arm_transpose_conv_s8_get_reverse_conv_buffer_size arm_transpose_conv_wrapper_s8

Fully-Connected Layer Functions

19 functions
arm_batch_matmul_s16 arm_batch_matmul_s8 arm_fully_connected_fp16 arm_fully_connected_per_channel_s16 arm_fully_connected_per_channel_s16_get_buffer_size arm_fully_connected_per_channel_s16_get_buffer_size_dsp arm_fully_connected_per_channel_s16_get_buffer_size_mve arm_fully_connected_per_channel_s8 arm_fully_connected_s16 arm_fully_connected_s16_get_buffer_size arm_fully_connected_s16_get_buffer_size_dsp arm_fully_connected_s16_get_buffer_size_mve arm_fully_connected_s4 arm_fully_connected_s8 arm_fully_connected_s8_get_buffer_size arm_fully_connected_s8_get_buffer_size_dsp arm_fully_connected_s8_get_buffer_size_mve arm_fully_connected_wrapper_s16 arm_fully_connected_wrapper_s8

Elementwise Functions

36 functions
arm_abs_s16 arm_abs_s8 arm_add_s16 arm_add_s8 arm_add_scalar_s16 arm_add_scalar_s8 arm_elementwise_add_s16 arm_elementwise_add_s8 arm_elementwise_mul_acc_s16 arm_elementwise_mul_s16 arm_elementwise_mul_s16_batch_offset arm_elementwise_mul_s16_s8 arm_elementwise_mul_s8 arm_elementwise_prelu_s8 arm_elementwise_squared_difference_s16 arm_elementwise_squared_difference_s8 arm_elementwise_sub_s16 arm_elementwise_sub_s8 arm_maximum_s16 arm_maximum_s8 arm_minimum_s16 arm_minimum_s8 arm_mul_s16 arm_mul_s8 arm_mul_scalar_s16 arm_mul_scalar_s8 arm_sqrt_s16 arm_sqrt_s8 arm_squared_difference_s16 arm_squared_difference_s8 arm_squared_difference_scalar_s16 arm_squared_difference_scalar_s8 arm_sub_s16 arm_sub_s8 arm_sub_scalar_s16 arm_sub_scalar_s8

Basic Math and Reduction

26 functions
arm_argmax_s16 arm_argmax_s8 arm_argmin_s16 arm_argmin_s8 arm_comparison_s16 arm_comparison_s8 arm_equal_s16 arm_equal_s8 arm_greater_equal_s16 arm_greater_equal_s8 arm_greater_s16 arm_greater_s8 arm_less_equal_s16 arm_less_equal_s8 arm_less_s16 arm_less_s8 arm_mean_s16 arm_mean_s8 arm_not_equal_s16 arm_not_equal_s8 arm_reduce_max_s16 arm_reduce_max_s8 arm_reduce_min_s16 arm_reduce_min_s8 arm_vector_sum_s8 arm_vector_sum_s8_s64

Activation Functions

19 functions
arm_clamp_s16 arm_clamp_s8 arm_hard_swish_compat_s8 arm_hard_swish_precise_s16 arm_hard_swish_precise_s8 arm_leaky_relu_s16 arm_leaky_relu_s8 arm_logistic_s16 arm_nn_activation_s16 arm_prelu_s8 arm_prelu_scalar_s8 arm_relu6_q7 arm_relu_generic_s16 arm_relu_generic_s8 arm_relu_q15 arm_relu_q7 arm_relu_s16 arm_relu_s8 arm_tanh_s16

Data Movement

31 functions
arm_batch_to_space_nd_s16 arm_batch_to_space_nd_s8 arm_concatenation_s16 arm_concatenation_s32 arm_concatenation_s8 arm_concatenation_s8_w arm_concatenation_s8_x arm_concatenation_s8_y arm_concatenation_s8_z arm_depth_to_space_s16 arm_depth_to_space_s8 arm_gather_nd_s16 arm_gather_nd_s8 arm_gather_s16 arm_gather_s8 arm_pad_s16 arm_pad_s8 arm_reshape_s8 arm_resize_nearest_neighbor_s16 arm_resize_nearest_neighbor_s8 arm_space_to_batch_nd_s16 arm_space_to_batch_nd_s8 arm_space_to_depth_s16 arm_space_to_depth_s8 arm_split_s16 arm_split_s8 arm_strided_slice_s16 arm_strided_slice_s32 arm_strided_slice_s8 arm_transpose_s16 arm_transpose_s8

Classifier Tail

21 functions
arm_avgpool_s16 arm_avgpool_s16_get_buffer_size arm_avgpool_s16_get_buffer_size_dsp arm_avgpool_s16_get_buffer_size_mve arm_avgpool_s8 arm_avgpool_s8_get_buffer_size arm_avgpool_s8_get_buffer_size_dsp arm_avgpool_s8_get_buffer_size_mve arm_dequantize_s16_f32 arm_dequantize_s8_f32 arm_max_pool_s16 arm_max_pool_s8 arm_q7_to_q15_with_offset arm_quantize_f32_s16 arm_quantize_f32_s8 arm_requantize_s16_s16 arm_requantize_s8_s8 arm_softmax_s16 arm_softmax_s8 arm_softmax_s8_s16 arm_softmax_u8

Sequence Functions

11 functions
arm_lstm_unidirectional_s16 arm_lstm_unidirectional_s8 arm_nn_lstm_calculate_gate_s16 arm_nn_lstm_calculate_gate_s8_s16 arm_nn_lstm_step_s16 arm_nn_lstm_step_s8 arm_svdf_s8 arm_svdf_s8_get_buffer_size arm_svdf_s8_get_buffer_size_dsp arm_svdf_s8_get_buffer_size_mve arm_svdf_state_s16_s8

Types and Support

Structs, enums, files, and lower-level support helpers are intentionally kept in the full generated C API index to avoid duplicating declarations across the grouped operator view and the complete Exhale tree.