Namespaces
	vector

Classes
struct	function_traits

class	CallHelper

class	NDArray
	NDArray is a class that can be used in both host and device code. NDArray is automatically divided into multiple NDArrayViews for each task. More...

struct	function_traits_helper

struct	function_traits_helper< R(*)(void)>

struct	function_traits_helper< R(*)(T1)>

struct	function_traits_helper< R(*)(T1, T2)>

struct	function_traits_helper< R(*)(T1, T2, T3)>

struct	function_traits_helper< R(*)(T1, T2, T3, T4)>

struct	function_traits_helper< R(*)(T1, T2, T3, T4, T5)>

struct	function_traits_helper< R(*)(T1, T2, T3, T4, T5, T6)>

struct	function_traits_helper< R(*)(T1, T2, T3, T4, T5, T6, T7)>

struct	function_traits_helper< R(*)(T1, T2, T3, T4, T5, T6, T7, T8)>

struct	function_traits_helper< R(*)(T1, T2, T3, T4, T5, T6, T7, T8, T9)>

struct	function_traits_helper< R(*)(T1, T2, T3, T4, T5, T6, T7, T8, T9, T10)>

class	remove_member_pointer

class	remove_member_pointer< R T::* >

struct	is_reference

struct	is_reference< T & >

struct	is_reference< T && >

struct	add_pointer

struct	remove_reference

struct	remove_reference< T & >

struct	remove_reference< T && >

struct	is_pointer

struct	is_pointer< T * >

struct	is_pointer< T *const >

struct	is_pointer< T *volatile >

struct	is_pointer< T *const volatile >

struct	remove_pointer

struct	remove_pointer< T * >

struct	remove_pointer< T *const >

struct	remove_pointer< T *volatile >

struct	remove_pointer< T *const volatile >

struct	is_ndarray

struct	is_ndarray< mu::NDArray< T > >

struct	KernelInfo

union	DpeOutput_t

union	DpeResult_t

union	DpeScore_t

union	DpeCsrDpb_t

union	DpeCsrLdb_t

Enumerations
enum class	DpeMode { Bsw = 0 , Mate = 1 , Global = 2 , DB = 3 , NumDpeMode = 3 }

enum class	DpeLaunch { Sync = 0 , Async = 1 }

enum class	DpeBuffer { DPB = 0 , LDB = 1 }

enum class	LogLevel { Trace , Debug , Info , Warn , Error , Critical , None }

Functions
void	Assert (bool cond, const char file=__builtin_FILE(), const char func=__builtin_FUNCTION(), int line=__builtin_LINE())
	Assertion function for runtime checks. More...

void	exit ()
	Terminates the MU kernel execution. More...

int32_t	atomicAdd (int32_t *address, int32_t val)
	Atomically adds a value to memory. More...

int32_t	atomicAddAsync (int32_t *address, int32_t val)
	Atomically adds a value to memory asynchronously. More...

uint32_t	atomicAnd (uint32_t *address, uint32_t val)
	Atomically performs bitwise AND operation with memory. More...

uint32_t	atomicAndAsync (uint32_t *address, uint32_t val)
	Atomically performs bitwise AND operation with memory asynchronously. More...

uint32_t	atomicOr (uint32_t *address, uint32_t val)
	Atomically performs bitwise OR operation with memory. More...

uint32_t	atomicOrAsync (uint32_t *address, uint32_t val)
	Atomically performs bitwise OR operation with memory asynchronously. More...

uint32_t	atomicXor (uint32_t *address, uint32_t val)
	Atomically performs bitwise XOR operation with memory. More...

uint32_t	atomicXorAsync (uint32_t *address, uint32_t val)
	Atomically performs bitwise XOR operation with memory asynchronously. More...

int32_t	atomicInc (int32_t *address, int32_t val)
	Atomically increments memory value. More...

int32_t	atomicIncAsync (int32_t *address, int32_t val)
	Atomically increments memory value asynchronously. More...

int32_t	atomicDec (int32_t *address, int32_t val)
	Atomically decrements memory value. More...

int32_t	atomicDecAsync (int32_t *address, int32_t val)
	Atomically decrements memory value asynchronously. More...

int32_t	atomicMin (int32_t *address, int32_t val)
	Atomically sets memory to minimum of current and new value. More...

int32_t	atomicMax (int32_t *address, int32_t val)
	Atomically sets memory to maximum of current and new value. More...

int32_t	atomicSwap (int32_t *address, int32_t val)
	Atomically exchanges memory value with new value. More...

int32_t	atomicCAS (int32_t *address, int32_t val, int32_t compare)
	Atomically compares and swaps memory value. More...

int64_t	atomicSwap58 (int64_t *address, int64_t val)
	Atomically exchanges memory value with new value (58-bit) More...

uint64_t	atomicOr58 (uint64_t *address, uint64_t val)
	Atomically performs bitwise OR operation with memory (58-bit) More...

uint64_t	atomicAnd58 (uint64_t *address, uint64_t val)
	Atomically performs bitwise AND operation with memory (58-bit) More...

uint64_t	atomicXor58 (uint64_t *address, uint64_t val)
	Atomically performs bitwise XOR operation with memory (58-bit) More...

int32_t	atomicLoad (int32_t *address)
	Atomically reads a value from memory. More...

int64_t	atomicLoad58 (int64_t *address)
	Atomically reads a value from memory (58-bit) More...

int32_t	atomicStoreAsync (int32_t *address, int32_t val)
	Atomically stores a value to memory asynchronously. More...

int64_t	atomicStore58Async (int64_t *address, int64_t val)
	Atomically stores a value to memory asynchronously (58-bit) More...

uint64_t	readGlobal (const void *src)
	Read a value from global cache. More...

void	writeGlobal (void *dst, uint64_t value)
	Write a value to global cache. More...

void	memcpyGlobal (void dst, const void src, uint64_t size)
	Copy data to global cache. More...

void	flushCache (const void *addr, uint64_t size=64)
	Flush cache lines for a memory region. More...

void	flushCacheAll ()
	Flush all cache lines for current thread. More...

void	prefetch (const void *addr, uint64_t size=64)
	Prefetch a contiguous memory region into cache. More...

void	prefetchStride (const void *addr, uint64_t count, uint64_t offset=64)
	Prefetch multiple cache lines at regular stride intervals. More...

bool	prefetchLinkedList (const void *addr, uint64_t count, uint64_t ptrOffset)
	Prefetch cache lines by traversing a linked list. More...

uint64_t	getClockCycle ()
	Get the current clock cycle count. More...

bool	isNormalMode (uint32_t mode)

	CALL_HELPER (0)

	CALL_HELPER (1)

	CALL_HELPER (2)

	CALL_HELPER (3)

	CALL_HELPER (4)

	CALL_HELPER (5)

	CALL_HELPER (6)

	CALL_HELPER (7)

	CALL_HELPER (8)

	CALL_HELPER (9)

template<typename FP >
void	callMain (FP func, uint32_t mode, uint32_t ndarrayValueBit, char *buffer)

void	muSetDpbMemory (uint32_t swId, uint8_t isSrc1Physical, uint64_t src1, uint8_t isSrc2Physical, uint64_t src2, uint32_t count)

void	muSetLdbMemory (uint32_t swId, uint8_t isSrc1Physical, uint64_t src1, uint8_t isSrc2Physical, uint64_t src2, uint32_t count)

uint64_t	muRunDpe (uint32_t swId, DpeMode mode, uint32_t tMask, uint32_t iLen, uint32_t jStart, uint32_t jLen, uint32_t weight, uint8_t isPhysical, uint64_t dstAddr, uint32_t tOffset, uint32_t tDelta, uint32_t maxStart, uint32_t skipVsb, DpeLaunch asyncMode)

uint64_t	getId ()
	Return the id of Mu. More...

uint64_t	getExecutionCount ()
	Return the execution count of Mu. More...

int	logPrintf (const char *format,...)
	Prints a formatted log message. More...

int	logPrintf (LogLevel level, const char *format,...)
	Prints a formatted log message with a specified log level. More...

void	asyncMessage (void *src, uint64_t size)
	Send asynchronous message to host. More...

void	syncMessage (void *src, uint64_t size)
	Send synchronous message to host and wait for response. More...

uint32_t	getTaskIdx (void)
	Get the current task index. More...

uint32_t	getTaskCount (void)
	Get the total number of tasks. More...

int	uartPrintf (const char *format,...)
	Print a formatted string to UART. More...

int	hostPrintf (const char *format,...)
	Print a formatted string to the host. More...

void	profilerStart (const char *tag)
	Start the profiler. More...

void	profilerEnd (const char *tag)
	End the profiler. More...

Variables
template<typename T >
constexpr bool	is_ndarray_v = is_ndarray<T>::value

constexpr uint64_t	DpeAddrAlignBit = 6

Enumeration Type Documentation

◆ DpeBuffer

enum mu::DpeBuffer

strong

Enumerator
DPB
LDB

Definition at line 27 of file dpe_common.hpp.

 {
     DPB = 0,
     LDB = 1,
 };

◆ DpeLaunch

enum mu::DpeLaunch

strong

Enumerator
Sync
Async

Definition at line 21 of file dpe_common.hpp.

 {
     Sync = 0,
     Async = 1
 };

◆ DpeMode

enum mu::DpeMode

strong

Enumerator
Bsw
Mate
Global
DB
NumDpeMode

Definition at line 12 of file dpe_common.hpp.

 {
     Bsw = 0,     // ksw_extend2
     Mate = 1,    // ksw_u8
     Global = 2,  // ksw_global2
     DB = 3,
     NumDpeMode = 3,
 };

◆ LogLevel

enum mu::LogLevel

strong

Enumerator
Trace
Debug
Info
Warn
Error
Critical
None

Definition at line 9 of file logger_const.hpp.

 {
     Trace,
     Debug,
     Info,
     Warn,
     Error,
     Critical,
     None,
 };

Function Documentation

◆ Assert()

void mu::Assert	(	bool	cond,
		const char *	file = `__builtin_FILE()`,
		const char *	func = `__builtin_FUNCTION()`,
		int	line = `__builtin_LINE()`
	)

Assertion function for runtime checks.

This function performs runtime assertion checks in debug builds. When the condition is false, it triggers a kernel halt for debugging.

Parameters

cond	The boolean condition to check
file	Source file where the assertion occurred (automatically filled)
func	Function name where the assertion occurred (automatically filled)
line	Line number where the assertion occurred (automatically filled)

◆ asyncMessage()

void mu::asyncMessage	(	void *	src,
		uint64_t	size
	)

Send asynchronous message to host.

Sends a message to the host system without waiting for a response. This function returns immediately after sending the message.

Parameters

src	Pointer to the source data to be sent
size	Size of the data in bytes

Note: This is a non-blocking operation

◆ atomicAdd()

int32_t mu::atomicAdd	(	int32_t *	address,
		int32_t	val
	)

Atomically adds a value to memory.

Computes (old + val) and stores the result in memory.

Parameters

address	Target memory address
val	Value to add

Returns: int32_t The old value before the operation

◆ atomicAddAsync()

int32_t mu::atomicAddAsync	(	int32_t *	address,
		int32_t	val
	)

Atomically adds a value to memory asynchronously.

Computes (old + val) and stores the result in memory.

Parameters

address	Target memory address
val	Value to add

Returns: int32_t Always returns 0

◆ atomicAnd()

uint32_t mu::atomicAnd	(	uint32_t *	address,
		uint32_t	val
	)

Atomically performs bitwise AND operation with memory.

Computes (old & val) and stores the result in memory.

Parameters

address	Target memory address
val	Value to AND with

Returns: uint32_t The old value before the operation

◆ atomicAnd58()

uint64_t mu::atomicAnd58	(	uint64_t *	address,
		uint64_t	val
	)

Atomically performs bitwise AND operation with memory (58-bit)

Note: Supported since MX1S Computes (old & val) and stores the result in memory

Parameters

address	Target memory address
val	Value to AND with

Returns: uint64_t The old value before the operation

◆ atomicAndAsync()

uint32_t mu::atomicAndAsync	(	uint32_t *	address,
		uint32_t	val
	)

Atomically performs bitwise AND operation with memory asynchronously.

Computes (old & val) and stores the result in memory.

Parameters

address	Target memory address
val	Value to AND with

Returns: uint32_t Always returns 0

◆ atomicCAS()

int32_t mu::atomicCAS	(	int32_t *	address,
		int32_t	val,
		int32_t	compare
	)

Atomically compares and swaps memory value.

Computes ((old == compare) ? val : old) and stores the result in memory

Parameters

address	Target memory address
val	New value to set if comparison succeeds
compare	Value to compare with

Returns: int32_t The old value before the operation

◆ atomicDec()

int32_t mu::atomicDec	(	int32_t *	address,
		int32_t	val
	)

Atomically decrements memory value.

Computes (((old == 0) || (old > val)) ? val : (old - 1)) and stores the result in memory.

Parameters

address	Target memory address
val	Minimum value before wrapping

Returns: int32_t The old value before the operation

◆ atomicDecAsync()

int32_t mu::atomicDecAsync	(	int32_t *	address,
		int32_t	val
	)

Atomically decrements memory value asynchronously.

Computes (((old == 0) || (old > val)) ? val : (old - 1)) and stores the result in memory.

Parameters

address	Target memory address
val	Minimum value before wrapping

Returns: int32_t Always returns 0

◆ atomicInc()

int32_t mu::atomicInc	(	int32_t *	address,
		int32_t	val
	)

Atomically increments memory value.

Computes ((old >= val) ? 0 : (old + 1)) and stores the result in memory.

Parameters

address	Target memory address
val	Maximum value before wrapping

Returns: int32_t The old value before the operation

◆ atomicIncAsync()

int32_t mu::atomicIncAsync	(	int32_t *	address,
		int32_t	val
	)

Atomically increments memory value asynchronously.

Computes ((old >= val) ? 0 : (old + 1)) and stores the result in memory.

Parameters

address	Target memory address
val	Maximum value before wrapping

Returns: int32_t Always returns 0

◆ atomicLoad()

int32_t mu::atomicLoad ( int32_t * address )

Atomically reads a value from memory.

Parameters

address Memory address to read from

Returns: int32_t The value read from the address

◆ atomicLoad58()

int64_t mu::atomicLoad58 ( int64_t * address )

Atomically reads a value from memory (58-bit)

Note: Supported since MX1S

Parameters

address Memory address to read from

Returns: uint64_t The value read from the address

◆ atomicMax()

int32_t mu::atomicMax	(	int32_t *	address,
		int32_t	val
	)

Atomically sets memory to maximum of current and new value.

Computes max(old, val) and stores the result in memory

Parameters

address	Target memory address
val	Value to compare with

Returns: int32_t The old value before the operation

◆ atomicMin()

int32_t mu::atomicMin	(	int32_t *	address,
		int32_t	val
	)

Atomically sets memory to minimum of current and new value.

Computes min(old, val) and stores the result in memory

Parameters

address	Target memory address
val	Value to compare with

Returns: int32_t The old value before the operation

◆ atomicOr()

uint32_t mu::atomicOr	(	uint32_t *	address,
		uint32_t	val
	)

Atomically performs bitwise OR operation with memory.

Computes (old | val) and stores the result in memory.

Parameters

address	Target memory address
val	Value to OR with

Returns: uint32_t The old value before the operation

◆ atomicOr58()

uint64_t mu::atomicOr58	(	uint64_t *	address,
		uint64_t	val
	)

Atomically performs bitwise OR operation with memory (58-bit)

Note: Supported since MX1S Computes (old | val) and stores the result in memory

Parameters

address	Target memory address
val	Value to OR with

Returns: uint64_t The old value before the operation

◆ atomicOrAsync()

uint32_t mu::atomicOrAsync	(	uint32_t *	address,
		uint32_t	val
	)

Atomically performs bitwise OR operation with memory asynchronously.

Computes (old | val) and stores the result in memory.

Parameters

address	Target memory address
val	Value to OR with

Returns: uint32_t Always returns 0

◆ atomicStore58Async()

int64_t mu::atomicStore58Async	(	int64_t *	address,
		int64_t	val
	)

Atomically stores a value to memory asynchronously (58-bit)

Stores a value to memory with atomic guarantees. This function performs the operation asynchronously.

Note: Supported since MX1S

Parameters

address	Target memory address
val	Value to store

Returns: int64_t Always returns 0

◆ atomicStoreAsync()

int32_t mu::atomicStoreAsync	(	int32_t *	address,
		int32_t	val
	)

Atomically stores a value to memory asynchronously.

Stores a value to memory with atomic guarantees. This function performs the operation asynchronously.

Note: Supported since MX1S

Parameters

address	Target memory address
val	Value to store

Returns: int32_t Always returns 0

◆ atomicSwap()

int32_t mu::atomicSwap	(	int32_t *	address,
		int32_t	val
	)

Atomically exchanges memory value with new value.

Parameters

address	Target memory address
val	Value to swap with

Returns: int32_t The old value before the operation

◆ atomicSwap58()

int64_t mu::atomicSwap58	(	int64_t *	address,
		int64_t	val
	)

Atomically exchanges memory value with new value (58-bit)

Note: Supported since MX1S

Parameters

address	Target memory address
val	Value to swap with

Returns: int64_t The old value before the operation

◆ atomicXor()

uint32_t mu::atomicXor	(	uint32_t *	address,
		uint32_t	val
	)

Atomically performs bitwise XOR operation with memory.

Computes (old ^ val) and stores the result in memory.

Parameters

address	Target memory address
val	Value to XOR with

Returns: uint32_t The old value before the operation

◆ atomicXor58()

uint64_t mu::atomicXor58	(	uint64_t *	address,
		uint64_t	val
	)

Atomically performs bitwise XOR operation with memory (58-bit)

Note: Supported since MX1S Computes (old ^ val) and stores the result in memory

Parameters

address	Target memory address
val	Value to XOR with

Returns: uint64_t The old value before the operation

◆ atomicXorAsync()

uint32_t mu::atomicXorAsync	(	uint32_t *	address,
		uint32_t	val
	)

Atomically performs bitwise XOR operation with memory asynchronously.

Computes (old ^ val) and stores the result in memory.

Parameters

address	Target memory address
val	Value to XOR with

Returns: uint32_t Always returns 0

◆ CALL_HELPER() [1/10]

mu::CALL_HELPER ( 0 )

◆ CALL_HELPER() [2/10]

mu::CALL_HELPER ( 1 )

◆ CALL_HELPER() [3/10]

mu::CALL_HELPER ( 2 )

◆ CALL_HELPER() [4/10]

mu::CALL_HELPER ( 3 )

◆ CALL_HELPER() [5/10]

mu::CALL_HELPER ( 4 )

◆ CALL_HELPER() [6/10]

mu::CALL_HELPER ( 5 )

◆ CALL_HELPER() [7/10]

mu::CALL_HELPER ( 6 )

◆ CALL_HELPER() [8/10]

mu::CALL_HELPER ( 7 )

◆ CALL_HELPER() [9/10]

mu::CALL_HELPER ( 8 )

◆ CALL_HELPER() [10/10]

mu::CALL_HELPER ( 9 )

◆ callMain()

template<typename FP >

void mu::callMain	(	FP	func,
		uint32_t	mode,
		uint32_t	ndarrayValueBit,
		char *	buffer
	)

inline

Definition at line 141 of file call_main.hpp.

 {
     CallHelper<typename remove_pointer<FP>::type>::invoke(func, mode, ndarrayValueBit, buffer);
 }

◆ exit()

void mu::exit ( )

Terminates the MU kernel execution.

Forces an immediate termination of the MU kernel's execution. This function is typically called after a critical assertion failure.

Note: This function does not return

◆ flushCache()

void mu::flushCache	(	const void *	addr,
		uint64_t	size = `64`
	)

Flush cache lines for a memory region.

This function flushes all cache lines that contain data from the specified memory region, ensuring that modified data in those cache lines is written back to memory. The function will flush all cache lines that overlap with the memory range [addr, addr + size).

Parameters

addr	Starting address of the memory region to flush
size	Size of the memory region in bytes (default: 64 bytes)

◆ flushCacheAll()

void mu::flushCacheAll ( )

Flush all cache lines for current thread.

This function flushes all cache lines in the cache hierarchy associated with the current thread, ensuring that all modified data across all cache lines is written back to memory. This operation only affects the cache lines belonging to the current thread's cache hierarchy.

◆ getClockCycle()

uint64_t mu::getClockCycle ( )

Get the current clock cycle count.

Returns the current processor clock cycle count.

MX1P: Returns 32-bit value (will overflow periodically)
MX1S: Returns 64-bit value

Note

For MX1P on a 1 GHz processor:

Each cycle is 1 nanosecond
2^32 cycles = ~4.29 seconds
Counter will overflow approximately every 4.29 seconds

Returns: uint64_t The current clock cycle count

◆ getExecutionCount()

uint64_t mu::getExecutionCount ( )

Return the execution count of Mu.

This function returns the execution count of Mu.

Returns: The execution count of Mu.

◆ getId()

uint64_t mu::getId ( )

Return the id of Mu.

This function returns the id of Mu.

Returns: The id of Mu.

◆ getTaskCount()

uint32_t mu::getTaskCount ( void )

Get the total number of tasks.

This function returns the total number of tasks in the current execution.

Returns: uint32_t The total number of tasks.

◆ getTaskIdx()

uint32_t mu::getTaskIdx ( void )

Get the current task index.

This function returns the index of the currently executing task.

Returns: uint32_t The index of the current task.

◆ hostPrintf()

int mu::hostPrintf	(	const char *	format,
			...
	)

Print a formatted string to the host.

This function sends a formatted string to the host interface. It behaves similarly to the standard printf function but targets the host.

Parameters

format	The format string.
...	Additional arguments specifying data to print.

Returns: int The number of characters printed.

◆ isNormalMode()

bool mu::isNormalMode ( uint32_t mode )

◆ logPrintf() [1/2]

int mu::logPrintf	(	const char *	format,
			...
	)

Prints a formatted log message.

This function prints a formatted log message using the specified format string and additional arguments. The log information is stored in memory and can be accessed via a CLI.

Parameters

format	The format string.
...	Additional arguments for the format string.

Returns: The number of characters printed.

◆ logPrintf() [2/2]

int mu::logPrintf	(	LogLevel	level,
		const char *	format,
			...
	)

Prints a formatted log message with a specified log level.

This function prints a formatted log message using the specified log level, format string, and additional arguments. The log information is stored in memory and can be accessed via a CLI.

Parameters

level	The log level.
format	The format string.
...	Additional arguments for the format string.

Returns: The number of characters printed.

◆ memcpyGlobal()

void mu::memcpyGlobal	(	void *	dst,
		const void *	src,
		uint64_t	size
	)

Copy data to global cache.

This function copies data from source to destination and writes it directly to the global cache for shared access across multiple cores.

Parameters

dst	Destination address in global cache
src	Source address to copy from
size	Number of bytes to copy

◆ muRunDpe()

uint64_t mu::muRunDpe	(	uint32_t	swId,
		DpeMode	mode,
		uint32_t	tMask,
		uint32_t	iLen,
		uint32_t	jStart,
		uint32_t	jLen,
		uint32_t	weight,
		uint8_t	isPhysical,
		uint64_t	dstAddr,
		uint32_t	tOffset,
		uint32_t	tDelta,
		uint32_t	maxStart,
		uint32_t	skipVsb,
		DpeLaunch	asyncMode
	)

◆ muSetDpbMemory()

void mu::muSetDpbMemory	(	uint32_t	swId,
		uint8_t	isSrc1Physical,
		uint64_t	src1,
		uint8_t	isSrc2Physical,
		uint64_t	src2,
		uint32_t	count
	)

◆ muSetLdbMemory()

void mu::muSetLdbMemory	(	uint32_t	swId,
		uint8_t	isSrc1Physical,
		uint64_t	src1,
		uint8_t	isSrc2Physical,
		uint64_t	src2,
		uint32_t	count
	)

◆ prefetch()

void mu::prefetch	(	const void *	addr,
		uint64_t	size = `64`
	)

Prefetch a contiguous memory region into cache.

This function prefetches memory starting from the specified address for the given size in bytes.

Parameters

addr	Starting address for prefetch. Should be 64-byte aligned (cache line boundary) for optimal performance.
size	Number of bytes to prefetch (valid range: 64 to 512 bytes, default: 64 bytes)

◆ prefetchLinkedList()

bool mu::prefetchLinkedList	(	const void *	addr,
		uint64_t	count,
		uint64_t	ptrOffset
	)

Prefetch cache lines by traversing a linked list.

This function prefetches cache lines by following a linked list structure. Starting from the given address, it reads the next pointer located at addr + ptrOffset and follows the chain, prefetching each node up to count times. The traversal stops early if a null pointer (0) is encountered at the next pointer location.

Parameters

addr	Starting address of the linked list node. Must be 8-byte aligned.
count	Maximum number of nodes to prefetch (valid range: 1 to 8)
ptrOffset	Offset to the next pointer in 8-byte units (valid range: 0 to 4088, must be 8-byte aligned)

Returns: true if prefetch succeeded, false if invalid parameters were provided

◆ prefetchStride()

void mu::prefetchStride	(	const void *	addr,
		uint64_t	count,
		uint64_t	offset = `64`
	)

Prefetch multiple cache lines at regular stride intervals.

This function prefetches cache lines starting from the specified address at regular intervals. The prefetch operation iterates from i = 0 to count - 1, prefetching the cache line at address: addr + offset * i.

Note: Only the cache line (64 bytes) at each calculated address is prefetched. Memory between stride intervals is NOT prefetched.

Parameters

addr	Starting address for prefetch. Should be 64-byte aligned (cache line boundary) for optimal performance.
count	Number of cache lines to prefetch (valid range: 1 to 8)
offset	Stride interval in bytes (valid range: 64 to 4088, must be 64-byte aligned, default: 64 bytes)

◆ profilerEnd()

void mu::profilerEnd ( const char * tag )

End the profiler.

This function ends the profiler.

◆ profilerStart()

void mu::profilerStart ( const char * tag )

Start the profiler.

This function starts the profiler.

◆ readGlobal()

uint64_t mu::readGlobal ( const void * src )

Read a value from global cache.

This function reads a 64-bit value from the specified address in the global cache. It ensures that the data is read directly from the global cache for shared access.

Parameters

src	Source address in global cache

Returns: uint64_t The value read from the global cache

◆ syncMessage()

void mu::syncMessage	(	void *	src,
		uint64_t	size
	)

Send synchronous message to host and wait for response.

Sends a message to the host system and blocks until receiving a response. The function will not return until the host processes the message.

Parameters

src	Pointer to the source data to be sent
size	Size of the data in bytes

Note: This is a blocking operation

See also: asyncMessage() for non-blocking alternative

◆ uartPrintf()

int mu::uartPrintf	(	const char *	format,
			...
	)

Print a formatted string to UART.

This function sends a formatted string to the UART interface. It behaves similarly to the standard printf function but targets UART.

Parameters

format	The format string.
...	Additional arguments specifying data to print.

Returns: int The number of characters printed.

◆ writeGlobal()

void mu::writeGlobal	(	void *	dst,
		uint64_t	value
	)

Write a value to global cache.

This function writes a 64-bit value to the specified address in the global cache. It ensures that the data is written directly to the global cache for shared access.

Parameters

dst	Destination address in global cache
value	Value to write

Variable Documentation

◆ DpeAddrAlignBit

constexpr uint64_t mu::DpeAddrAlignBit = 6

constexpr

Definition at line 11 of file dpe_common.hpp.

◆ is_ndarray_v

template<typename T >

constexpr bool mu::is_ndarray_v = is_ndarray<T>::value

constexpr

Definition at line 264 of file call_main_template.hpp.

Namespaces

Classes

Enumerations

Functions

Variables

Enumeration Type Documentation

◆ DpeBuffer

◆ DpeLaunch

◆ DpeMode

◆ LogLevel

Function Documentation

◆ Assert()

◆ asyncMessage()

◆ atomicAdd()

◆ atomicAddAsync()

◆ atomicAnd()

◆ atomicAnd58()

◆ atomicAndAsync()

◆ atomicCAS()

◆ atomicDec()

◆ atomicDecAsync()

◆ atomicInc()

◆ atomicIncAsync()

◆ atomicLoad()

◆ atomicLoad58()

◆ atomicMax()

◆ atomicMin()

◆ atomicOr()

◆ atomicOr58()

◆ atomicOrAsync()

◆ atomicStore58Async()

◆ atomicStoreAsync()

◆ atomicSwap()

◆ atomicSwap58()

◆ atomicXor()

◆ atomicXor58()

◆ atomicXorAsync()

◆ CALL_HELPER() [1/10]

◆ CALL_HELPER() [2/10]

◆ CALL_HELPER() [3/10]

◆ CALL_HELPER() [4/10]

◆ CALL_HELPER() [5/10]

◆ CALL_HELPER() [6/10]

◆ CALL_HELPER() [7/10]

◆ CALL_HELPER() [8/10]

◆ CALL_HELPER() [9/10]

◆ CALL_HELPER() [10/10]

◆ callMain()

◆ exit()

◆ flushCache()

◆ flushCacheAll()

◆ getClockCycle()

◆ getExecutionCount()

◆ getId()

◆ getTaskCount()

◆ getTaskIdx()

◆ hostPrintf()

◆ isNormalMode()

◆ logPrintf() [1/2]

◆ logPrintf() [2/2]

◆ memcpyGlobal()

◆ muRunDpe()

◆ muSetDpbMemory()

◆ muSetLdbMemory()

◆ prefetch()

◆ prefetchLinkedList()

◆ prefetchStride()

◆ profilerEnd()

◆ profilerStart()

◆ readGlobal()

◆ syncMessage()

◆ uartPrintf()

◆ writeGlobal()

Variable Documentation

◆ DpeAddrAlignBit

◆ is_ndarray_v