Kernel APIs Reference

Global Functions

xilLz4Compress

#include "lz4_compress_mm.hpp"

void xilLz4Compress (
    const xf::compression::uintMemWidth_t* in,
    xf::compression::uintMemWidth_t* out,
    uint32_t* compressd_size,
    uint32_t* in_block_size,
    uint32_t block_size_in_kb,
    uint32_t input_size
    )

LZ4 compression kernel takes the raw data as input and compresses the data in block based fashion and writes the output to global memory.

Parameters:

in	input raw data
out	output compressed data
compressd_size	compressed output size of each block
in_block_size	input block size of each block
block_size_in_kb	input block size in bytes
input_size	input data size

xilLz4CompressStream

#include "lz4_compress_stream.hpp"

void xilLz4CompressStream (
    hls::stream <ap_axiu <8, 0, 0, 0>>& inaxistream,
    hls::stream <ap_axiu <8, 0, 0, 0>>& outaxistream,
    uint32_t inputSize
    )

LZ4 compression streaming kernel. It takes input from axi kernel stream and writes compressed data back to output axi kernel stream.

Parameters:

inaxistream	Input axi kernel stream
outaxistream	Output axi kernel stream
inputSize	Input compressed data size

xilLz4Decompress

#include "lz4_decompress_mm.hpp"

void xilLz4Decompress (
    const xf::compression::uintMemWidth_t* in,
    xf::compression::uintMemWidth_t* out,
    uint32_t* in_block_size,
    uint32_t* in_compress_size,
    uint32_t block_size_in_kb,
    uint32_t no_blocks
    )

LZ4 decompression kernel takes compressed data as input and process in block based fashion and writes the raw data to global memory.

Parameters:

in	input compressed data
out	output raw data
in_block_size	input block size of each block
in_compress_size	compress size of each block
block_size_in_kb	block size in bytes
no_blocks	number of blocks

xilLz4DecompressStream

#include "lz4_decompress_stream.hpp"

void xilLz4DecompressStream (
    hls::stream <ap_axiu <8, 0, 0, 0>>& inaxistream,
    hls::stream <ap_axiu <8, 0, 0, 0>>& outaxistream,
    uint32_t inputSize,
    uint32_t outputSize
    )

LZ4 decompression streaming kernel. It takes input from axi kernel stream and writes decompressed data back to output axi kernel stream.

Parameters:

inaxistream	Input axi kernel stream
outaxistream	Output axi kernel stream
inputSize	Input compressed data size
outputSize	Output decompressed data size

xilLz4Decompress

#include "lz4_multibyte_decompress_mm.hpp"

void xilLz4Decompress (
    const ap_uint <PARALLEL_BYTE*8>* in,
    ap_uint <PARALLEL_BYTE*8>* out,
    uint32_t* in_block_size,
    uint32_t* in_compress_size,
    uint32_t block_size_in_kb,
    uint32_t no_blocks
    )

LZ4 decompression kernel takes compressed data as input and process in block based fashion and writes the raw data to global memory.

Parameters:

in	input compressed data
out	output raw data
in_block_size	input block size of each block
in_compress_size	compress size of each block
block_size_in_kb	block size in bytes
no_blocks	number of blocks

xilLz4P2PDecompress

#include "lz4_p2p_decompress_kernel.hpp"

void xilLz4P2PDecompress (
    const xf::compression::uintMemWidth_t* in,
    xf::compression::uintMemWidth_t* out,
    dt_blockInfo* bObj,
    dt_chunkInfo* cObj,
    uint32_t block_size_in_kb,
    uint32_t compute_unit,
    uint8_t total_no_cu,
    uint32_t num_blocks
    )

LZ4 P2P decompression kernel is responsible for decompressing data which is in LZ4 encoded form.

Parameters:

in	input stream width
out	output stream width
in_block_size	input size
in_compress_size	output size
block_start_idx	start index of block
no_blocks	number of blocks for each compute unit
block_size_in_kb	block input size
compute_unit	particular compute unit
total_no_cu	number of compute units
num_blocks	number of blocks base don host buffersize

xilLz4Packer

#include "lz4_packer_mm.hpp"

void xilLz4Packer (
    uint512_t* in,
    uint512_t* out,
    uint32_t* compressd_size,
    uint32_t* in_block_size,
    uint32_t* encoded_size,
    uint512_t* orig_input_data,
    uint32_t block_size_in_kb,
    uint32_t no_blocks,
    uint32_t xxhashVal,
    uint32_t input_size
    )

LZ4 packer kernel takes the raw data as input and compresses the data in block based fashion and writes the output to global memory.

Parameters:

in	input raw data
out	output compressed data
compressd_size	compressed output size of each block
in_block_size	input block size of each block
encoded_size	encoded size of each block
orig_input_data	raw input data
block_size_in_kb	input block size in bytes
no_blocks	number of input blocks
xxhashVal	Hash Value
input_size	Total Input File Size

xilLz4Unpacker

#include "lz4_unpacker_kernel.hpp"

void xilLz4Unpacker (
    const xf::compression::uintMemWidth_t* in,
    dt_blockInfo* bObj,
    dt_chunkInfo* cObj,
    uint32_t block_size_in_kb,
    uint8_t first_chunk,
    uint8_t total_no_cu,
    uint32_t num_blocks
    )

LZ4 unpacker kernel is responsible in unpacking LZ4 compressed block information.

Parameters:

in	input stream width
in_block_size	input block size
in_compress_size	input compress size
block_start_idx	start index of each input block
no_blocks_per_cu	number of blocks for each compute unit
original_size	original file size
in_start_index	input start index
no_blocks	number of blocks
block_size_in_kb	size of each block
first_chunk	first chunk to determine header
total_no_cu	number of decompress compute units
num_blocks	number of blocks based on host buffersize

xilSnappyCompress

#include "snappy_compress_mm.hpp"

void xilSnappyCompress (
    const xf::compression::uintMemWidth_t* in,
    xf::compression::uintMemWidth_t* out,
    uint32_t* compressd_size,
    uint32_t* in_block_size,
    uint32_t block_size_in_kb,
    uint32_t input_size
    )

Snappy compression kernel takes the raw data as input and compresses the data in block based fashion and writes the output to global memory.

Parameters:

in	input raw data
out	output compressed data
compressd_size	compressed output size of each block
in_block_size	input block size of each block
block_size_in_kb	input block size in bytes
input_size	input data size

xilSnappyCompressStream

#include "snappy_compress_stream.hpp"

void xilSnappyCompressStream (
    hls::stream <ap_axiu <8, 0, 0, 0>>& inaxistream,
    hls::stream <ap_axiu <8, 0, 0, 0>>& outaxistream,
    uint32_t inputSize
    )

Snappy compression streaming kernel takes the raw data as input from kernel axi stream and compresses the data in block based fashion and writes the output to kernel axi stream.

Parameters:

inaxistream	input kernel axi stream for raw data
outaxistream	output kernel axi stream for compressed data
inputSize	input data size

xilSnappyDecompress

#include "snappy_decompress_mm.hpp"

void xilSnappyDecompress (
    const xf::compression::uintMemWidth_t* in,
    xf::compression::uintMemWidth_t* out,
    uint32_t* in_block_size,
    uint32_t* in_compress_size,
    uint32_t block_size_in_kb,
    uint32_t no_blocks
    )

Snappy decompression kernel takes compressed data as input and process in block based fashion and writes the raw data to global memory.

Parameters:

in	input compressed data
out	output raw data
in_block_size	input block size of each block
in_compress_size	compress size of each block
block_size_in_kb	block size in bytes
no_blocks	number of blocks

xilSnappyDecompressStream

#include "snappy_decompress_stream.hpp"

void xilSnappyDecompressStream (
    hls::stream <ap_axiu <8, 0, 0, 0>>& inaxistream,
    hls::stream <ap_axiu <8, 0, 0, 0>>& outaxistream,
    uint32_t inputSize,
    uint32_t outputSize
    )

Snappy decompression streaming kernel takes compressed data as input from kernel axi stream and process in block based fashion and writes the raw data to global memory.

Parameters:

inaxistream	input kernel axi stream for compressed data
outaxistream	output kernel axi stream for decompressed data
inputSize	input data size

xilSnappyDecompress

#include "snappy_multibyte_decompress_mm.hpp"

void xilSnappyDecompress (
    const xf::compression::uintMemWidth_t* in,
    xf::compression::uintMemWidth_t* out,
    uint32_t* in_block_size,
    uint32_t* in_compress_size,
    uint32_t block_size_in_kb,
    uint32_t no_blocks
    )

Snappy decompression kernel takes compressed data as input and process in block based fashion and writes the raw data to global memory.

Parameters:

in	input compressed data
out	output raw data
in_block_size	input block size of each block
in_compress_size	compress size of each block
block_size_in_kb	block size in bytes
no_blocks	number of blocks

xilSnappyDecompressStream

#include "snappy_multibyte_decompress_stream.hpp"

void xilSnappyDecompressStream (
    hls::stream <ap_axiu <MULTIPLE_BYTES*8, 0, 0, 0>>& inaxistream,
    hls::stream <ap_axiu <MULTIPLE_BYTES*8, 0, 0, 0>>& outaxistream,
    hls::stream <ap_axiu <32, 0, 0, 0>>& outaxistreamsize,
    uint32_t inputSize
    )

Snappy decompression streaming kernel takes compressed data as input from kernel axi stream and process in block based fashion and writes the raw data to global memory.

Parameters:

inaxistream	input kernel axi stream for compressed data
outaxistream	output kernel axi stream for decompressed data
inputSize	input data size

xilDecompressStream

#include "zlib_decompress_stream.hpp"

void xilDecompressStream (
    uint32_t input_size,
    hls::stream <ap_axiu <kGMemDWidth, 0, 0, 0>>& inaxistreamd,
    hls::stream <ap_axiu <kGMemDWidth, 0, 0, 0>>& outaxistreamd,
    hls::stream <ap_axiu <64, 0, 0, 0>>& sizestreamd
    )

Zlib decompression stream kernel top function.

Parameters:

input_size	input size
inaxistreamd	input kernel axi stream
outaxistreamd	output kernel axi stream

xilHuffmanKernel

#include "zlib_huffman_enc_mm.hpp"

void xilHuffmanKernel (
    xf::compression::uintMemWidth_t* in,
    uint32_t* lit_freq,
    uint32_t* dist_freq,
    xf::compression::uintMemWidth_t* out,
    uint32_t* in_block_size,
    uint32_t* compressd_size,
    uint32_t block_size_in_kb,
    uint32_t input_size
    )

Huffman kernel top function. This is an initial version of Huffman Kernel which does block based bit packing process. It uses dynamic huffman codes and bit lengths to encode the LZ77 (Byte Compressed Data) output. This version operates on 1MB block data per engine as this is suitable for use cases where raw data is over >100MB and compression ratio is over 2.5x in order to achieve best throughput. This can be further optimized to achieve better throughput for smaller file usecase.

Parameters:

in	input stream
out	output stream
in_block_size	input block size
compressd_size	output compressed size
dyn_litmtree_codes	input literal and match length codes
dyn_distree_codes	input distance codes
dyn_bitlentree_codes	input bit-length codes
dyn_litmtree_blen	input literal and match length bit length data
dyn_dtree_blen	input distance bit length data
dyn_bitlentree_blen	input bit-length of bit length data
dyn_max_codes	input maximum codes
block_size_in_kb	input block size in bytes
input_size	input data size

xilLz77Compress

#include "zlib_lz77_compress_mm.hpp"

void xilLz77Compress (
    const xf::compression::uintMemWidth_t* in,
    xf::compression::uintMemWidth_t* out,
    uint32_t* compressd_size,
    uint32_t* in_block_size,
    uint32_t* dyn_ltree_freq,
    uint32_t* dyn_dtree_freq,
    uint32_t block_size_in_kb,
    uint32_t input_size
    )

LZ77 compression kernel takes the raw data as input and compresses the data in block based fashion and writes the output to global memory. LZ77 is a byte based compression scheme. The resulting output from this kernel is represented in packet form of 32bit length <Literal, Match Length, Distance>. It also generates output of literal and distance frequencies for dynamic huffman tree generation. The output generated by this kernel is referred by TreeGen and Huffman Kernels.

Parameters:

in	input stream
out	output stream
compressd_size	compressed output size of each block
in_block_size	input block size of each block
dyn_ltree_freq	literal frequency data
dyn_dtree_freq	distance frequency data
block_size_in_kb	input block size in bytes
input_size	input data size

xilTreegenKernel

#include "zlib_treegen_mm.hpp"

void xilTreegenKernel (
    uint32_t* dyn_ltree_freq,
    uint32_t* dyn_dtree_freq,
    uint32_t* dyn_bltree_freq,
    uint32_t* dyn_ltree_codes,
    uint32_t* dyn_dtree_codes,
    uint32_t* dyn_bltree_codes,
    uint32_t* dyn_ltree_blen,
    uint32_t* dyn_dtree_blen,
    uint32_t* dyn_bltree_blen,
    uint32_t* max_codes,
    uint32_t block_size_in_kb,
    uint32_t input_size,
    uint32_t blocks_per_chunk
    )

This is a resource optimized version of huffman treegen kernel. It takes literal and distance frequency data as input through single input stream and generates dynamic huffman codes and bit length data which is output through a single output stream. This kernel does not use DDR in any way and is optimised for both speed and low resource usage.

Parameters:

freqStream	24-bit input stream for getting frequency data
codeStream	20-bit output stream sending huffman codes and bit-lengths data