Miscellaneous

Read lfsr

Access to LFSR register which is used to generate random numbers. Once the register is read, it randomizes a new number.

Intrinsics Prototype and Arguments

char256 read_lfsr_b(int switches=0, char256 income={}, bool predicate=1, bool polarity=0)

char256 read_lfsr_vb(int switches, char256 income, bool256 predicate, bool polarity=0)

switches

Switches of LFSR instructions.

income

This value is returned if the predicate is false.

predicate

Predicate value for the instruction.

polarity

True if polarity of the predicate is inverted.

Allowed switches are:

  • SW_READ_ONLY - Don’t reseed value.

A value of LFSR (VPU_LFSR).

Write lfsr

Write to LFSR register which is used to generate random numbers.

Intrinsics Prototype and Arguments

void write_lfsr_b(char256 src, bool predicate=1, bool polarity=0)

void write_lfsr_vb(char256 src, bool256 predicate, bool polarity=0)

src

The argument (SRC1).

predicate

Predicate value for the instruction.

polarity

True if polarity of the predicate is inverted.

Read lane id

Access to a LANE_ID_* register.

Intrinsics Prototype and Arguments

uint64 read_lane_id_4b_b(uint64 income={}, bool predicate=1, bool polarity=0)

uint64 read_lane_id_4b_vb(uint64 income, bool64 predicate, bool polarity=0)

ushort128 read_lane_id_2b_b(ushort128 income={}, bool predicate=1, bool polarity=0)

ushort128 read_lane_id_2b_vb(ushort128 income, bool128 predicate, bool polarity=0)

uchar256 read_lane_id_1b_b(uchar256 income={}, bool predicate=1, bool polarity=0)

uchar256 read_lane_id_1b_vb(uchar256 income, bool256 predicate, bool polarity=0)

income

This value is returned if the predicate is false.

predicate

Predicate value for the instruction.

polarity

True if polarity of the predicate is inverted.

A value of LANE_ID_*.

Get dim size

Return dimension size for the given tensor id.

Intrinsics Prototype and Arguments

uint32_t get_dim_size(int32_t a, uint32_t dim)

a

Tensor id.

dim

Dimension.

Dimension size.

Set dim size

Set dimension size for the given tensor id.

Intrinsics Prototype and Arguments

void set_dim_size(int32_t a, uint32_t dim, uint32_t value)

a

Tensor id.

dim

Dimension.

value

The new value of size.

Get dim stride

Return dimension stride for the given tensor id.

Intrinsics Prototype and Arguments

uint32_t get_dim_stride(int32_t a, uint32_t dim)

a

Tensor id.

dim

Dimension.

Dimension stride.

Set dim size

Set dimension stride for the given tensor id.

Intrinsics Prototype and Arguments

void set_dim_stride(int32_t a, uint32_t dim, uint32_t value)

a

Tensor id.

dim

Dimension.

value

The new value of stride.

Get pad value

Return padding value for the given tensor id.

Intrinsics Prototype and Arguments

uint32_t get_pad_value_uint(int32_t a)

int32_t get_pad_value_int(int32_t a)

float get_pad_value_float(int32_t a)

uint16_t get_pad_value_ushort(int32_t a)

int16_t get_pad_value_short(int32_t a)

uint8_t get_pad_value_uchar(int32_t a)

int8_t get_pad_value_char(int32_t a)

bf16 get_pad_value_bf16(int32_t a)

a

Tensor id.

Padding value.

Set pad value

Set padding value for the given tensor id.

Intrinsics Prototype and Arguments

void set_pad_value_uint(int32_t a, uint32_t value)

void set_pad_value_int(int32_t a, int32_t value)

void set_pad_value_float(int32_t a, float value)

void set_pad_value_ushort(int32_t a, uint16_t value)

void set_pad_value_short(int32_t a, int16_t value)

void set_pad_value_uchar(int32_t a, uint8_t value)

void set_pad_value_char(int32_t a, int8_t value)

void set_pad_value_bf16(int32_t a, bf16 value)

a

Tensor id.

value

The new padding value.