c7100/FFTLIB_types.h File Reference

Introduction

File to hold common structure, enums, macros and functions for FFTLIB.

Version

0.1 - May 2020 : Initial Version

Definition in file c7100/FFTLIB_types.h.

Go to the source code of this file.

Macros
#define	FFTLIB_BYTE_WIDTH   64
	MMA width in bytes. More...
#define	FFTLIB_CALC_STRIDE(BYTES, ALIGN_SHIFT)    (((((BYTES)-1) >> (ALIGN_SHIFT)) + 1) << (ALIGN_SHIFT))
	Macro function that computes the minimum stride for the specified number of bytes and alignment. More...
#define	FFTLIB_MMA_SIZE_8_BIT_SHIFT    6
	Macro for FFTLIB shift amounts. More...
#define	FFTLIB_MMA_SIZE_16_BIT_SHIFT    5
	Shift for MMA size when data type is 16-bit integers. More...
#define	FFTLIB_MMA_SIZE_32_BIT_SHIFT    4
	Shift for MMA size when data type is 32-bit integers. More...
#define	FFTLIB_MMA_SIZE_8_BIT
	MMA size as a function of precision. More...
#define	FFTLIB_MMA_SIZE_16_BIT
	type is 16-bit integers More...
#define	FFTLIB_MMA_SIZE_32_BIT
	type is 32-bit integers More...

Typedefs
typedef FFTLIB_quantMethod_type_e	FFTLIB_quantMethod_type

Enumerations
enum	FFTLIB_activation_type_e {   FFTLIB_NONE = 0 , FFTLIB_RELU , FFTLIB_SATURATION , FFTLIB_ROUNDING ,   FFTLIB_NONE = 0 , FFTLIB_RELU , FFTLIB_SATURATION , FFTLIB_ROUNDING ,   FFTLIB_PSAT , FFTLIB_NONE = 0 , FFTLIB_RELU , FFTLIB_SATURATION ,   FFTLIB_ROUNDING , FFTLIB_PSAT }
	The enumeration of possible activation operations on the computation. More...
enum	FFTLIB_quantMethod_type_e {   FFTLIB_QUANT_PER_KERNEL = 0 , FFTLIB_QUANT_PER_CHANNEL = 1 , FFTLIB_QUANT_PER_KERNEL = 0 , FFTLIB_QUANT_PER_CHANNEL = 1 ,   FFTLIB_QUANT_PER_KERNEL = 0 , FFTLIB_QUANT_PER_CHANNEL = 1 }

Macro Definition Documentation

FFTLIB_BYTE_WIDTH

#define FFTLIB_BYTE_WIDTH   64

MMA width in bytes.

Definition at line 74 of file c7100/FFTLIB_types.h.

FFTLIB_CALC_STRIDE

#define FFTLIB_CALC_STRIDE	(		BYTES,
			ALIGN_SHIFT
	)		(((((BYTES)-1) >> (ALIGN_SHIFT)) + 1) << (ALIGN_SHIFT))

Macro function that computes the minimum stride for the specified number of bytes and alignment.

Definition at line 83 of file c7100/FFTLIB_types.h.

FFTLIB_MMA_SIZE_8_BIT_SHIFT

#define FFTLIB_MMA_SIZE_8_BIT_SHIFT    6

Macro for FFTLIB shift amounts.

Shift for MMA size when data type is 8-bit integers

Definition at line 88 of file c7100/FFTLIB_types.h.

FFTLIB_MMA_SIZE_16_BIT_SHIFT

#define FFTLIB_MMA_SIZE_16_BIT_SHIFT    5

Shift for MMA size when data type is 16-bit integers.

Definition at line 90 of file c7100/FFTLIB_types.h.

FFTLIB_MMA_SIZE_32_BIT_SHIFT

#define FFTLIB_MMA_SIZE_32_BIT_SHIFT    4

Shift for MMA size when data type is 32-bit integers.

Definition at line 92 of file c7100/FFTLIB_types.h.

FFTLIB_MMA_SIZE_8_BIT

#define FFTLIB_MMA_SIZE_8_BIT

Value:

  ((uint32_t)(((uint32_t)1)                                                     \
             << FFTLIB_MMA_SIZE_8_BIT_SHIFT))

MMA size as a function of precision.

Shift for MMA size when data type is 8-bit integers

Definition at line 99 of file c7100/FFTLIB_types.h.

FFTLIB_MMA_SIZE_16_BIT

#define FFTLIB_MMA_SIZE_16_BIT

Value:

  ((uint32_t)(((uint32_t)1)                                                     \
             << FFTLIB_MMA_SIZE_16_BIT_SHIFT))

type is 16-bit integers

Shift for MMA size when data

Definition at line 103 of file c7100/FFTLIB_types.h.

FFTLIB_MMA_SIZE_32_BIT

#define FFTLIB_MMA_SIZE_32_BIT

Value:

  ((int32_t)(((uint32_t)1)                                                     \
             << FFTLIB_MMA_SIZE_32_BIT_SHIFT))

type is 32-bit integers

Shift for MMA size when data

Definition at line 107 of file c7100/FFTLIB_types.h.

Typedef Documentation

FFTLIB_quantMethod_type

typedef FFTLIB_quantMethod_type_e FFTLIB_quantMethod_type

Definition at line 66 of file c7100/FFTLIB_types.h.

Enumeration Type Documentation

FFTLIB_activation_type_e

enum FFTLIB_activation_type_e

The enumeration of possible activation operations on the computation.

Enumerator
FFTLIB_NONE	Return result as-is (allow wrapping)
FFTLIB_RELU	Return result after applying ReLU operation.
FFTLIB_SATURATION	Return result after saturating.
FFTLIB_ROUNDING	Return result after rounding.
FFTLIB_NONE	Return result as-is (allow wrapping)
FFTLIB_RELU	Return result after applying ReLU operation.
FFTLIB_SATURATION	Return result after saturating.
FFTLIB_ROUNDING	Return result after rounding.
FFTLIB_PSAT	Return result after parameterized saturation.
FFTLIB_NONE	Return result as-is (allow wrapping)
FFTLIB_RELU	Return result after applying ReLU operation.
FFTLIB_SATURATION	Return result after saturating.
FFTLIB_ROUNDING	Return result after rounding.
FFTLIB_PSAT	Return result after parameterized saturation.

Definition at line 55 of file c7100/FFTLIB_types.h.

FFTLIB_quantMethod_type_e

enum FFTLIB_quantMethod_type_e

Enumerator
FFTLIB_QUANT_PER_KERNEL
FFTLIB_QUANT_PER_CHANNEL
FFTLIB_QUANT_PER_KERNEL
FFTLIB_QUANT_PER_CHANNEL
FFTLIB_QUANT_PER_KERNEL
FFTLIB_QUANT_PER_CHANNEL

Definition at line 62 of file c7100/FFTLIB_types.h.