NN Softmax Functions

The softmax functions are exponential functions with base 2. More...

Functions
static void	hpm_nn_softmax_s8_fast (const q7_t *in_vec, const uint16_t size, q7_t *out_vec)
	This is a softmax function for signed 8-bit integer input vectors. More...
static void	hpm_nn_softmax_s16_fast (const q15_t *in_vec, const uint16_t size, q15_t *out_vec)
	This is a softmax function for signed 16-bit integer input vectors. More...
static void	hpm_nn_softmax_s8_hp (const int8_t *in_tensor, const int32_t in_tensor_row, const int32_t in_tensor_col, const int32_t scale, const int32_t lshift, const int32_t diff_min, int8_t *out_tensor)
	This is a softmax function for signed 8-bit integer input tensor with high precision algorithm. More...
static void	hpm_nn_softmax_u8_hp (const uint8_t *in_tensor, const int32_t in_tensor_row, const int32_t in_tensor_col, const int32_t scale, const int32_t lshift, const int32_t diff_min, uint8_t *out_tensor)
	This is a softmax function for unsigned 8-bit integer input tensor with high precision algorithm. More...

Detailed Description

The softmax functions are exponential functions with base 2.

Function Documentation

hpm_nn_softmax_s16_fast()

static void hpm_nn_softmax_s16_fast ( const q15_t *  in_vec, const uint16_t  size, q15_t *  out_vec  )

inlinestatic

#include <middleware/hpm_math/hpm_math.h>

This is a softmax function for signed 16-bit integer input vectors.

Parameters

[in]	in_vec	pointer of the input vector
[in]	size	number of elements in the input vector
[out]	out_vec	pointer of the output vector

hpm_nn_softmax_s8_fast()

static void hpm_nn_softmax_s8_fast ( const q7_t *  in_vec, const uint16_t  size, q7_t *  out_vec  )

inlinestatic

#include <middleware/hpm_math/hpm_math.h>

This is a softmax function for signed 8-bit integer input vectors.

Parameters

[in]	in_vec	pointer of the input vector
[in]	size	number of elements in the input vector
[out]	out_vec	pointer of the output vector

Example:

#define LENGTH 10
q7_t in_data[LENGTH] = {...};
q7_t out_data[LENGTH];
 
hpm_nn_softmax_s8_fast(in_data, LENGTH, out_data);

hpm_nn_softmax_s8_hp()

static void hpm_nn_softmax_s8_hp ( const int8_t *  in_tensor, const int32_t  in_tensor_row, const int32_t  in_tensor_col, const int32_t  scale, const int32_t  lshift, const int32_t  diff_min, int8_t *  out_tensor  )

inlinestatic

#include <middleware/hpm_math/hpm_math.h>

This is a softmax function for signed 8-bit integer input tensor with high precision algorithm.

Parameters

[in]	in_tensor	pointer of the input tensor
[in]	in_tensor_row	number of rows in the input tensor
[in]	in_tensor_col	number of columns in the input tensor
[in]	scale	scaling value for input quantization
[in]	lshift	left shift amount for input quantization
[in]	diff_min	minimum threshold to perform the quantized exponential operation. The difference can be obtained by subtracting the input from the maximum in row.
[out]	out_tensor	pointer of the output tensor

hpm_nn_softmax_u8_hp()

static void hpm_nn_softmax_u8_hp ( const uint8_t *  in_tensor, const int32_t  in_tensor_row, const int32_t  in_tensor_col, const int32_t  scale, const int32_t  lshift, const int32_t  diff_min, uint8_t *  out_tensor  )

inlinestatic

#include <middleware/hpm_math/hpm_math.h>

This is a softmax function for unsigned 8-bit integer input tensor with high precision algorithm.

Parameters

[in]	in_tensor	pointer of the input tensor
[in]	in_tensor_row	number of rows in the input tensor
[in]	in_tensor_col	number of columns in the input tensor
[in]	scale	scaling value for input quantization
[in]	lshift	left shift amount for input quantization
[in]	diff_min	minimum threshold to perform the quantized exponential operation. The difference can be obtained by subtracting the input from the maximum in row.
[out]	out_tensor	pointer of the output tensor