ref: 6a9831a6b038638266165dc1e9f115678f0b330e
parent: 1ddfcfd48cb87f8dc29240d705a4da78bae0eb50
author: Jean-Marc Valin <jmvalin@jmvalin.ca>
date: Thu Jan 18 13:16:54 EST 2024
Remove run-time code for old TF2 models No longer needed now that PLC is trained with PyTorch stack
--- a/dnn/nnet.c
+++ b/dnn/nnet.c
@@ -115,78 +115,6 @@
}
}
-void _lpcnet_compute_dense(const DenseLayer *layer, float *output, const float *input, int arch)
-{- LinearLayer matrix;
- celt_assert(input != output);
- matrix.bias = layer->bias;
- matrix.subias = NULL;
- matrix.float_weights = layer->input_weights;
- matrix.weights = NULL;
- matrix.weights_idx = NULL;
- matrix.diag = NULL;
- matrix.nb_inputs = layer->nb_inputs;
- matrix.nb_outputs = layer->nb_neurons;
- matrix.scale = NULL;
- compute_linear(&matrix, output, input, arch);
- compute_activation(output, output, layer->nb_neurons, layer->activation, arch);
-}
-
-#ifdef USE_SU_BIAS
-#define bias_type subias
-#else
-#define bias_type bias
-#endif
-#define MAX_IDX_SIZE 8192
-
-void compute_gruB(const GRULayer *gru, const float* gru_b_condition, float *state, const float *input, int arch)
-{- LinearLayer in_matrix, rec_matrix;
- int i, M, N;
- float bias[3*MAX_RNN_NEURONS_ALL];
- float scale[3*MAX_RNN_NEURONS_ALL];
- M = gru->nb_inputs;
- N = gru->nb_neurons;
-
- in_matrix.bias = bias;
- in_matrix.diag = NULL;
- in_matrix.nb_inputs = M;
- in_matrix.nb_outputs = 3*N;
- in_matrix.subias = bias;
-#ifdef DISABLE_DOT_PROD
- for (i=0;i<3*N;i++) bias[i] = gru->bias[i] + gru_b_condition[i];
- in_matrix.scale = NULL;
- in_matrix.float_weights = gru->input_weights;
- in_matrix.weights = NULL;
-#else
- for (i=0;i<3*N;i++) bias[i] = gru->bias_type[i] + gru_b_condition[i];
- for (i=0;i<3*N;i++) scale[i] = SCALE_1;
- in_matrix.scale = scale;
- in_matrix.weights = gru->input_weights;
- in_matrix.float_weights = NULL;
-#endif
- in_matrix.weights_idx = gru->input_weights_idx;
-
- rec_matrix.bias = &gru->bias[3*N];
- rec_matrix.diag = NULL;
- rec_matrix.nb_inputs = N;
- rec_matrix.nb_outputs = 3*N;
- rec_matrix.scale = scale;
- rec_matrix.subias = &gru->subias[3*N];
-#ifdef DISABLE_DOT_PROD
- rec_matrix.scale = NULL;
- rec_matrix.float_weights = gru->recurrent_weights;
- rec_matrix.weights = NULL;
-#else
- rec_matrix.scale = scale;
- rec_matrix.weights = gru->recurrent_weights;
- rec_matrix.float_weights = NULL;
-#endif
- rec_matrix.weights_idx = NULL;
- compute_generic_gru(&in_matrix, &rec_matrix, state, input, arch);
-}
-
-
#define MAX_CONV_INPUTS_ALL DRED_MAX_CONV_INPUTS
void compute_generic_conv1d(const LinearLayer *layer, float *output, float *mem, const float *input, int input_size, int activation, int arch)
--- a/dnn/nnet.h
+++ b/dnn/nnet.h
@@ -31,13 +31,6 @@
#include <stddef.h>
#include "opus_types.h"
-#ifdef DISABLE_DOT_PROD
-typedef float qweight;
-#else
-typedef signed char qweight;
-#define DOT_PROD
-#endif
-
#define ACTIVATION_LINEAR 0
#define ACTIVATION_SIGMOID 1
#define ACTIVATION_TANH 2
@@ -91,41 +84,7 @@
int kheight;
} Conv2dLayer;
-typedef struct {- const float *bias;
- const float *input_weights;
- int nb_inputs;
- int nb_neurons;
- int activation;
-} DenseLayer;
-typedef struct {- const float *bias;
- const float *subias;
- const qweight *input_weights;
- const int *input_weights_idx;
- const qweight *recurrent_weights;
- int nb_inputs;
- int nb_neurons;
- int activation;
- int reset_after;
-} GRULayer;
-
-typedef struct {- const float *bias;
- const float *input_weights;
- int nb_inputs;
- int kernel_size;
- int nb_neurons;
- int activation;
-} Conv1DLayer;
-
-typedef struct {- const float *embedding_weights;
- int nb_inputs;
- int dim;
-} EmbeddingLayer;
-
void compute_generic_dense(const LinearLayer *layer, float *output, const float *input, int activation, int arch);
void compute_generic_gru(const LinearLayer *input_weights, const LinearLayer *recurrent_weights, float *state, const float *in, int arch);
void compute_generic_conv1d(const LinearLayer *layer, float *output, float *mem, const float *input, int input_size, int activation, int arch);
@@ -134,10 +93,6 @@
void compute_gated_activation(const LinearLayer *layer, float *output, const float *input, int activation, int arch);
-void _lpcnet_compute_dense(const DenseLayer *layer, float *output, const float *input, int arch);
-
-void compute_gruB(const GRULayer *gru, const float* gru_b_condition, float *state, const float *input, int arch);
-
int parse_weights(WeightArray **list, const unsigned char *data, int len);
@@ -168,24 +123,6 @@
int out_channels,
int ktime,
int kheight);
-
-int dense_init(DenseLayer *layer, const WeightArray *arrays,
- const char *bias,
- const char *input_weights,
- int nb_inputs,
- int nb_neurons,
- int activation);
-
-int gru_init(GRULayer *layer, const WeightArray *arrays,
- const char *bias,
- const char *subias,
- const char *input_weights,
- const char *input_weights_idx,
- const char *recurrent_weights,
- int nb_inputs,
- int nb_neurons,
- int activation,
- int reset_after);
void compute_linear_c(const LinearLayer *linear, float *out, const float *in);
--- a/dnn/parse_lpcnet_weights.c
+++ b/dnn/parse_lpcnet_weights.c
@@ -176,46 +176,6 @@
return 0;
}
-
-int dense_init(DenseLayer *layer, const WeightArray *arrays,
- const char *bias,
- const char *input_weights,
- int nb_inputs,
- int nb_neurons,
- int activation)
-{- if ((layer->bias = find_array_check(arrays, bias, nb_neurons*sizeof(layer->bias[0]))) == NULL) return 1;
- if ((layer->input_weights = find_array_check(arrays, input_weights, nb_inputs*nb_neurons*sizeof(layer->input_weights[0]))) == NULL) return 1;
- layer->nb_inputs = nb_inputs;
- layer->nb_neurons = nb_neurons;
- layer->activation = activation;
- return 0;
-}
-
-int gru_init(GRULayer *layer, const WeightArray *arrays,
- const char *bias,
- const char *subias,
- const char *input_weights,
- const char *input_weights_idx,
- const char *recurrent_weights,
- int nb_inputs,
- int nb_neurons,
- int activation,
- int reset_after)
-{- int total_blocks;
- if ((layer->bias = find_array_check(arrays, bias, 6*nb_neurons*sizeof(layer->bias[0]))) == NULL) return 1;
- if ((layer->subias = find_array_check(arrays, subias, 6*nb_neurons*sizeof(layer->subias[0]))) == NULL) return 1;
- if ((layer->input_weights_idx = find_idx_check(arrays, input_weights_idx, nb_inputs, 3*nb_neurons, &total_blocks)) == NULL) return 1;
- if ((layer->input_weights = find_array_check(arrays, input_weights, SPARSE_BLOCK_SIZE*total_blocks*sizeof(layer->input_weights[0]))) == NULL) return 1;
- if ((layer->recurrent_weights = find_array_check(arrays, recurrent_weights, 3*nb_neurons*nb_neurons*sizeof(layer->recurrent_weights[0]))) == NULL) return 1;
- layer->nb_inputs = nb_inputs;
- layer->nb_neurons = nb_neurons;
- layer->activation = activation;
- layer->reset_after = reset_after;
- return 0;
-}
-
int conv2d_init(Conv2dLayer *layer, const WeightArray *arrays,
const char *bias,
const char *float_weights,
--
⑨