Class that represents an array of data records in the profile buffer. More...

#include <gt_profile_array.h>

Public Member Functions
	GtProfileArray ()
	Default constructor.
	GtProfileArray (uint32_t recordSize, uint32_t numRecords, uint32_t numThreadBuckets=1, uint32_t alignment=64)
	Constructor.
bool	Allocate (IGtProfileBufferAllocator &allocator)
	Reserve space required to store all chunks of this array in the profile buffer.
bool	Read (const IGtProfileBuffer &profileBuffer, void *records, uint32_t fistRecordNum, uint32_t numRecords=1, uint32_t threadBucketNum=0) const
	Read data from the specified records of this array in the profile buffer.
bool	Write (IGtProfileBuffer &profileBuffer, const void *records, uint32_t fistRecordNum, uint32_t numRecords=1, uint32_t threadBucketNum=0) const
	Write data into specified records of this array in the profile buffer (see the Read() parameters descriptions)
bool	Initialize (IGtProfileBuffer &profileBuffer) const
	Zero-initialize memory area allocated for this array in the specified profile buffer.
void	ComputeOffset (const IGtGenCoder &coder, IGtGenProcedure &proc, const GtReg &offsetReg, uint32_t recordNum=0, uint32_t offsetInRecord=0) const
	Generate code that computes offset/address within the specified record.
void	ComputeAddress (const IGtGenCoder &coder, IGtGenProcedure &proc, const GtReg &addrReg, uint32_t recordNum=0, uint32_t offsetInRecord=0) const
void	ComputeOffset (const IGtGenCoder &coder, IGtGenProcedure &proc, const GtReg &offsetReg, const GtReg &offsetInChunkReg, uint32_t offsetInRecord=0) const
void	ComputeAddress (const IGtGenCoder &coder, IGtGenProcedure &proc, const GtReg &addrReg, const GtReg &offsetInChunkReg, uint32_t offsetInRecord=0) const
void	ComputeRelAddress (const IGtGenCoder &coder, IGtGenProcedure &proc, const GtReg &addrReg, const GtReg &baseReg, int32_t offset) const
	Generate code that computes address of a (field) location within a record relative to the specified base address within the same record.
uint32_t	RecordOffset (uint32_t recordNum, uint32_t threadBucketNum) const
	Get offset of the specified record in the specified thread bucket, relative to the profile buffer base.
uint32_t	Size () const
	Return size of the memory block, in bytes.
GtProfileBufferOffset	Offset () const
	Return offset of the allocated memory block, or -1.
bool	IsAllocated () const
	Return true if this array has been allocated.
uint32_t	RecordSize () const
	Return size of the record, in bytes.
uint32_t	NumRecords () const
	Return number of data records in the array.
uint32_t	NumThreadBuckets () const
	Return number of thread buckets (chunks)
uint32_t	ChunkSize () const
	Return size of the per-bucket chunk, in bytes.

Detailed Description

Class that represents an array of data records in the profile buffer.

The data can be collected in a single global array or distributed among multiple thread buckets each of which stores profile data in its own array (chunk)

Terminology:

GTPin groups threads in buckets to minimize contention of concurrent memory accesses. Thread bucket is a group of threads that store their profile data in the same, shared memory block of the profile buffer. Threads that belong to different buckets use separate, not interleaving memory areas of the profile buffer, called chunks.
The number of thread buckets (memory chunks) is a configurable value, determined by the tool. The valid range of this value is [1, max. number of HW threads]. To simplify code generation, GTPin requires number of thread buckets to be power-of-two.
To map a thread into the corresponding thread bucket, GTPin applies some hash function to the thread ID.

Availability:

Constructor & Destructor Documentation

gtpin::GtProfileArray::GtProfileArray	(	uint32_t	recordSize,
		uint32_t	numRecords,
		uint32_t	numThreadBuckets = `1`,
		uint32_t	alignment = `64`
	)		`[inline]`

Constructor.

Parameters:

[in]	recordSize	Size of the record in bytes
[in]	numRecords	Number of data records in the array
[in]	numThreadBuckets	Number of thread buckets each of which will collect profile data in its own array, called chunk (see 'Terminology' section). The value of this parameter always rounded up to the next power of two.
[in]	alignment	Alignment of the array in the profile buffer; must be power of two

Member Function Documentation

bool gtpin::GtProfileArray::Allocate ( IGtProfileBufferAllocator & allocator ) [inline]

Reserve space required to store all chunks of this array in the profile buffer.

Parameters:

[in] allocator Allocator to be used to reserve space

Returns:: true - success, false - failure

Precondition:: !IsAllocated()

void gtpin::GtProfileArray::ComputeAddress	(	const IGtGenCoder &	coder,
		IGtGenProcedure &	proc,
		const GtReg &	addrReg,
		const GtReg &	offsetInChunkReg,
		uint32_t	offsetInRecord = `0`
	)		const `[inline]`

Generate code that computes offset/address within the specified record.

Generate code that computes offset/address within a record that resides in the memory chunk of the currently running HW thread. The offset is relative to the profile buffer base.

Note:: If both recordNum = 0 and offsetInRecord = 0, the generated code computes offset/address of the memory chunk associated with the currently running HW thread.

Parameters:

[in]	coder	GEN code generator
[in]	proc	Procedure, the generated code is appended to
[in]	offsetReg	Register that receives the calculated offset within the profile buffer
[in]	addrReg	Register that receives the calculated address
[in]	recordNum	Ordinal number of the record
[in]	offsetInChunkReg	32-bit register that holds offset (of the record) relative to the chunk base
[in]	offsetInRecord	Offset within the record

Precondition:: IsAllocated() && (recordNum < NumRecords()) && (offsetInRecord < RecordSize())

void gtpin::GtProfileArray::ComputeAddress	(	const IGtGenCoder &	coder,
		IGtGenProcedure &	proc,
		const GtReg &	addrReg,
		uint32_t	recordNum = `0`,
		uint32_t	offsetInRecord = `0`
	)		const `[inline]`

Generate code that computes offset/address within the specified record.

Generate code that computes offset/address within a record that resides in the memory chunk of the currently running HW thread. The offset is relative to the profile buffer base.

Note:: If both recordNum = 0 and offsetInRecord = 0, the generated code computes offset/address of the memory chunk associated with the currently running HW thread.

Parameters:

[in]	coder	GEN code generator
[in]	proc	Procedure, the generated code is appended to
[in]	offsetReg	Register that receives the calculated offset within the profile buffer
[in]	addrReg	Register that receives the calculated address
[in]	recordNum	Ordinal number of the record
[in]	offsetInChunkReg	32-bit register that holds offset (of the record) relative to the chunk base
[in]	offsetInRecord	Offset within the record

Precondition:: IsAllocated() && (recordNum < NumRecords()) && (offsetInRecord < RecordSize())

void gtpin::GtProfileArray::ComputeOffset	(	const IGtGenCoder &	coder,
		IGtGenProcedure &	proc,
		const GtReg &	offsetReg,
		uint32_t	recordNum = `0`,
		uint32_t	offsetInRecord = `0`
	)		const `[inline]`

Generate code that computes offset/address within the specified record.

Generate code that computes offset/address within a record that resides in the memory chunk of the currently running HW thread. The offset is relative to the profile buffer base.

Note:: If both recordNum = 0 and offsetInRecord = 0, the generated code computes offset/address of the memory chunk associated with the currently running HW thread.

Parameters:

[in]	coder	GEN code generator
[in]	proc	Procedure, the generated code is appended to
[in]	offsetReg	Register that receives the calculated offset within the profile buffer
[in]	addrReg	Register that receives the calculated address
[in]	recordNum	Ordinal number of the record
[in]	offsetInChunkReg	32-bit register that holds offset (of the record) relative to the chunk base
[in]	offsetInRecord	Offset within the record

Precondition:: IsAllocated() && (recordNum < NumRecords()) && (offsetInRecord < RecordSize())

void gtpin::GtProfileArray::ComputeOffset	(	const IGtGenCoder &	coder,
		IGtGenProcedure &	proc,
		const GtReg &	offsetReg,
		const GtReg &	offsetInChunkReg,
		uint32_t	offsetInRecord = `0`
	)		const `[inline]`

Generate code that computes offset/address within the specified record.

Generate code that computes offset/address within a record that resides in the memory chunk of the currently running HW thread. The offset is relative to the profile buffer base.

Note:: If both recordNum = 0 and offsetInRecord = 0, the generated code computes offset/address of the memory chunk associated with the currently running HW thread.

Parameters:

[in]	coder	GEN code generator
[in]	proc	Procedure, the generated code is appended to
[in]	offsetReg	Register that receives the calculated offset within the profile buffer
[in]	addrReg	Register that receives the calculated address
[in]	recordNum	Ordinal number of the record
[in]	offsetInChunkReg	32-bit register that holds offset (of the record) relative to the chunk base
[in]	offsetInRecord	Offset within the record

Precondition:: IsAllocated() && (recordNum < NumRecords()) && (offsetInRecord < RecordSize())

void gtpin::GtProfileArray::ComputeRelAddress	(	const IGtGenCoder &	coder,
		IGtGenProcedure &	proc,
		const GtReg &	addrReg,
		const GtReg &	baseReg,
		int32_t	offset
	)		const `[inline]`

Generate code that computes address of a (field) location within a record relative to the specified base address within the same record.

Parameters:

[in]	coder	GEN code generator
[in]	proc	Procedure, the generated code is appended to
[in]	addrReg	Register that receives the calculated address
[in]	baseReg	Register that holds an address within the record
[in]	offset	Offset within the record, relative to 'baseReg'

bool gtpin::GtProfileArray::Initialize ( IGtProfileBuffer & profileBuffer ) const [inline]

Zero-initialize memory area allocated for this array in the specified profile buffer.

Parameters:

[in] profileBuffer Profile buffer in which this array is allocated

Returns:: true - success, false - failure

Precondition:: IsAllocated()

bool gtpin::GtProfileArray::Read	(	const IGtProfileBuffer &	profileBuffer,
		void *	records,
		uint32_t	fistRecordNum,
		uint32_t	numRecords = `1`,
		uint32_t	threadBucketNum = `0`
	)		const `[inline]`

Read data from the specified records of this array in the profile buffer.

Parameters:

[in]	profileBuffer	Profile buffer to read records from
[in]	records	Pointer to an array that receives records from the profile buffer. The size of this array should be at least (numRecords * RecordSize()) bytes.
[in]	threadBucketNum	Ordinal number of the thread bucket (memory chunk) to read records from
[in]	fistRecordNum	Ordinal number of the first record to read
[in]	numRecords	Number of records to read

Returns:: true - success, false - failure

Precondition:: IsAllocated() && (threadBucketNum < NumThreadBuckets()) && (fistRecordNum + numRecords <= NumRecords())

uint32_t gtpin::GtProfileArray::RecordOffset	(	uint32_t	recordNum,
		uint32_t	threadBucketNum
	)		const `[inline]`

Get offset of the specified record in the specified thread bucket, relative to the profile buffer base.

Parameters:

[in]	recordNum	Ordinal number of the record
[in]	threadBucketNum	Ordinal number of the thread bucket

Returns:: Offset of the specified record

Precondition:: IsAllocated() && (threadBucketNum < NumThreadBuckets()) && (recordNum < NumRecords())

bool gtpin::GtProfileArray::Write	(	IGtProfileBuffer &	profileBuffer,
		const void *	records,
		uint32_t	fistRecordNum,
		uint32_t	numRecords = `1`,
		uint32_t	threadBucketNum = `0`
	)		const `[inline]`

Write data into specified records of this array in the profile buffer (see the Read() parameters descriptions)

Returns:: true - success, false - failure

Precondition:: IsAllocated() && (threadBucketNum < NumThreadBuckets()) && (fistRecordNum + numRecords <= NumRecords())

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation