Software Acceleration Classes¶

enum xf::data_analytics::text::re::ErrCode¶

Overview¶

Error code enumerations. More…

// enum values

SUCCESS = 0
PATTERN_ERR = 1
MEM_ERR = 2
DEV_ERR = 3
OPCODE_ERR = 4
INSTR_NM_ERR = 5
CCLASS_NM_ERR = 6
CPGP_NM_ERR = 7

Detailed Documentation¶

Error code enumerations.

Enum Values¶

SUCCESS

RE pre-compilation success

PATTERN_ERR

Invalid RE pattern

MEM_ERR

CL buffer allocation failed

DEV_ERR

Kernel creation failed

OPCODE_ERR

Wrong OP code

INSTR_NM_ERR

Wrong number of instructions

CCLASS_NM_ERR

Wrong number of character classes

CPGP_NM_ERR

Wrong number of capturing groups

class xf::data_analytics::text::re::RegexEngine¶

#include "xf_data_analytics/text/regex_engine.hpp"

Overview¶

Offload regex match with FPGA.

Methods¶

RegexEngine¶

RegexEngine (
    const std::string& xclbin,
    const int dev_index = 0,
    const int instr_depth = 4096,
    const int char_class_num = 128,
    const int capture_grp_num = 512,
    const int msg_size = 4096,
    const int max_slice_size = 5242880,
    const int max_slice_num = 256
    )

Default constructor for loading and programming binary.

Parameters:

xclbin	Path to FPGA binary.
dev_index	The index of Xilinx OpenCL device.
instr_depth	Max number of instructions.
char_class_num	Max number of character classes.
capture_grp_num	Max number of capturing groups.
msg_size	Max size for each message, in number of bytes. max_slice_size Max message slice size, in number of bytes.
max_slice_num	Max message slice number.

~RegexEngine¶

~RegexEngine ()

Default de-constructor for releasing program/command-queue/contect.

compile¶

ErrCode compile (std::string pattern)

Pre-compiles pattern and gives error code correspondingly.

Parameters:

pattern

Input regular expression

getCpgpNm¶

uint32_t getCpgpNm () const

Gets the number of capturing groups for current pattern.

match¶

ErrCode match (
    uint32_t total_lnm,
    const uint64_t* msg_buff,
    uint32_t* offt_buff,
    uint16_t* len_buff,
    uint32_t* out_buff
    )

Prepares configurations for RegexEngineKernel , and performs the actual matching process.

Parameters:

total_lnm	Total number of lines in current message block
msg_buff	Buffer for saving messages line-by-line, pads white-space if not hit the boundary of 64-bit
offt_buff	Buffer for saving starting address of each line, aligned with 64-bit
len_buff	Buffer for saving length of each message, in byte
out_buff	Buffer for saving output results including match flag & corresponding start/end offset address for each capturing group

RegexEngine¶

RegexEngine (
    const int instr_depth = 4096,
    const int char_class_num = 128,
    const int capture_grp_num = 512,
    const int msg_size = 4096,
    const int max_slice_size = 5242880,
    const int max_slice_num = 256
    )

Default constructor for loading and programming binary.

Parameters:

instr_depth	Max number of instructions.
char_class_num	Max number of character classes.
capture_grp_num	Max number of capturing groups.
msg_size	Max size for each message, in number of bytes. max_slice_size Max message slice size, in number of bytes.
max_slice_num	Max message slice number.

class sssd_engine::DataEngineConfig¶

#include "data_engine_config.hpp"

Overview¶

generating configuration bits for CSV scanner kernel

Methods¶

DataEngineConfig¶

DataEngineConfig ()

default constructor

genConfigBits¶

ErrorCode genConfigBits (
    bool gzip,
    sssd_scandesc_t* sd,
    uint64_t* cfg
    )

generate configuration bits for kernel

Parameters:

gzip	file under gzip or not
sd	schema for describing the table
cfg	kernel configurations

Returns:

error code

class sssd_engine::data_engine_sc::DataEngine¶

#include "data_engine_sc.hpp"

Overview¶

class to manage kernels on single device

// fields

std::thread run_t

Fields¶

std::thread run_t

thread for run data engine;

Methods¶

DataEngine¶

DataEngine (
    int t_id,
    VPP_BP& _csvInBufPool,
    VPP_BP& _cfgInBufPool,
    VPP_BP& _outBufPool,
    VPP_BP& _metaBufPool,
    VPP_CC* _cuCluster
    )

constructor of data engine. context, program, command queue are created and ready after FPGA init

Parameters:

t_id	targeted device id.
_csvInBufPool	input CSV buffer pool
_cfgInBufPool	input configuration buffer pool
_outBufPool	output result buffer pool
_metaBufPool	output meta buffer pool
_cuCluster	CU clusters

~DataEngine¶

~DataEngine ()

default deconstructor

pushRequest¶

void pushRequest (
    std::promise <RetObj> prom,
    std::string file_path,
    size_t size,
    uint64_t* cfg
    )

push request to queue

Parameters:

prom	promise to synchronize status of execution
file_path	file path
size	file size
cfg	pointer to configuration buffer

release¶

void release (char* buf)

release buffer

Parameters:

buf	buffer to be released

class sssd_engine::SmartSSDCache¶

#include "smart_ssd_cache.hpp"

Overview¶

class to handle a specific SSD device

Methods¶

getCardNum¶

int getCardNum ()

get number of cards

Returns:

number of cards

SmartSSDCache¶

SmartSSDCache (
    const char* xclbin_path,
    int card_num,
    sssd_info_t* disks,
    FILE* log
    )

default constructor, instantiate data engine

Parameters:

xclbin_path	path of xclbin
card_num	number of availabe SmartSSD cards
disks	information of the SSDs
log	file handler for dumping the logs

~SmartSSDCache¶

~SmartSSDCache ()

default deconstructor

addFile¶

ErrorCode addFile (
    std::unique_ptr <char[]> data,
    size_t size,
    const std::string& filename
    )

store file to managed SmartSSD cards. Copy data from CPU memory to SSD cards

Parameters:

data	pointer to file content in memory. Before return, the block of storage will be released
size	file size
filename	file to store

Returns:

success or fail status

scanFile¶

char* scanFile (
    const char* fname,
    sssd_scandesc_t* sd,
    ErrorCode& err
    )

scan files from managed SmartSSD cards.

Parameters:

fname	file name;
sd	pointer to configuration including schema, projection and filter conditions filenames.
err	return the status of exectution.

Returns:

pointer the selected row. It is stored in format defined by data type in schema.

listFiles¶

ErrorCode listFiles (
    const char* path_pattern,
    std::vector <std::string>& file_list
    )

list all the file name that matche the path_pattern

Parameters:

path_pattern	specify the path_pattern for regex
file_list	store the file name that matched the path_pattern

Returns:

fail/success

print_input¶

void print_input (const sssd_scandesc_t* sd)

print input information

Parameters:

sd	the input schema

print_output¶

void print_output (
    const int64_t value [],
    const bool isnull [],
    int32_t hash,
    const sssd_scandesc_t* sd
    )

print output information

Parameters:

value	values
isnull	is null flags
hash	hash value
sd	the schema

release¶

void release (
    const char* file_path,
    char* buf_ptr
    )

release specified buffer

Parameters:

file_path	path to the file
buf_ptr	buffer to be released