#include <assert.h>
#include <math.h>
#include <stddef.h>
#include <stdint.h>
#include <string.h>
#include "xnnpack.h"
#include "xnnpack/allocator.h"
#include "xnnpack/cache.h"
#include "xnnpack/common.h"
#include "xnnpack/compute.h"
#include "xnnpack/config.h"
#include "xnnpack/log.h"
#include "xnnpack/math.h"
#include "xnnpack/microfnptr.h"
#include "xnnpack/microkernel-type.h"
#include "xnnpack/microparams-init.h"
#include "xnnpack/microparams.h"
#include "xnnpack/operator-type.h"
#include "xnnpack/operator-utils.h"
#include "xnnpack/operator.h"
#include "xnnpack/pack.h"
#include "xnnpack/params.h"
#include "pthreadpool.h"
enum xnn_status create_batch_matrix_multiply_nc(
uint32_t flags,
const void* params,
size_t params_size,
const struct xnn_gemm_config* gemm_config,
const struct gemm_fused_ukernels* gemm_ukernels,
xnn_packw_gemm_gio_ukernel_fn pack_gemm_gio,
enum xnn_operator_type operator_type,
xnn_operator_t* batch_matrix_multiply_op_out)
{ … }
enum xnn_status xnn_create_batch_matrix_multiply_nc_f32(
uint32_t flags,
xnn_operator_t* batch_matrix_multiply_op_out)
{ … }
enum xnn_status xnn_create_batch_matrix_multiply_nc_f16(
uint32_t flags,
xnn_operator_t* batch_matrix_multiply_op_out)
{ … }
enum xnn_status xnn_create_batch_matrix_multiply_nc_qd8_f32_qc8w(
size_t batch_size_b, size_t k, size_t n, const int8_t* data_b,
const float* scale_b, uint32_t flags,
xnn_operator_t* batch_matrix_multiply_op_out) { … }
static enum xnn_status reshape_batch_matrix_multiply_nc(
xnn_operator_t batch_matrix_multiply_op,
enum xnn_operator_type expected_operator_type, size_t num_batch_dims,
const size_t* batch_dims_a, const size_t* batch_dims_b, size_t m, size_t k,
size_t n, size_t* workspace_size, size_t* workspace_alignment,
uint32_t log2_input_a_element_size, uint32_t log2_input_b_element_size,
uint32_t bias_element_size, uint32_t w_stride_extra_bytes,
uint32_t log2_output_element_size, const void* params, size_t params_size,
size_t num_threads) { … }
enum xnn_status xnn_reshape_batch_matrix_multiply_nc_f16(
xnn_operator_t batch_matrix_multiply_op, size_t num_batch_dims,
const size_t* batch_dims_a, const size_t* batch_dims_b, size_t m, size_t k,
size_t n, size_t* workspace_size, size_t* workspace_alignment,
pthreadpool_t threadpool) { … }
enum xnn_status xnn_reshape_batch_matrix_multiply_nc_f32(
xnn_operator_t batch_matrix_multiply_op, size_t num_batch_dims,
const size_t* batch_dims_a, const size_t* batch_dims_b, size_t m, size_t k,
size_t n, size_t* workspace_size, size_t* workspace_alignment,
pthreadpool_t threadpool) { … }
enum xnn_status xnn_reshape_batch_matrix_multiply_nc_qd8_f32_qc8w(
xnn_operator_t batch_matrix_multiply_op, size_t num_batch_dims,
const size_t* batch_dims_a, const size_t* batch_dims_b, size_t m, size_t k,
size_t n, pthreadpool_t threadpool) { … }
static enum xnn_status setup_batch_matrix_multiply_nc(
xnn_operator_t batch_matrix_multiply_op,
enum xnn_operator_type expected_operator_type, const void* input_a,
const struct xnn_dynamic_quantization_params* quantization_params,
const void* input_b, void* packed_weights, void* output) { … }
enum xnn_status xnn_setup_batch_matrix_multiply_nc_f16(
xnn_operator_t batch_matrix_multiply_op, void* workspace,
const void* input_a, const void* input_b, void* output) { … }
enum xnn_status xnn_setup_batch_matrix_multiply_nc_f32(
xnn_operator_t batch_matrix_multiply_op, void* workspace,
const float* input_a, const float* input_b, float* output) { … }
enum xnn_status xnn_setup_batch_matrix_multiply_nc_qd8_f32_qc8w(
xnn_operator_t batch_matrix_multiply_op, const int8_t* input_a,
const struct xnn_dynamic_quantization_params* quantization_params,
float* output) { … }