Check out the bzip2 functions from Intel IPP 5.2 beta

[Ying_S_Intel]

Dear Customers,
I'd like to ask in this forum if anyone has a chance to check out the new functions we developed for bzip2 functions.

If you are familiar with the bzip2 algorithm, it is well known for the high quality on compression. With the Intel IPP 5.2 beta, we introduced 12 new APIs for bzip2 functions as below:

bzip2:
Main primitives:
ippsDecodeHuff_BZ2_8u16u
ippsunpackHuffcontect_BZ2_8u16u
ippsEncodeHuff_BZ2_16u8u
ippsPackHuffContext_BZ2_16u8u

Service primitives:
ippsDecodeHuffGetSize_BZ2_8u16u
ippsDecodeHuffInitAlloc_BZ2_8u16u
ippsDecodeHuffInit_BZ2_8u16u ippsDecodeHuffFree_BZ2_8u16u
ippsEncodeHuffGetSize_BZ2_16u8u
ippsEncodeHuffInitAlloc_BZ2_16u8u
ippsEncodeHuffInit_BZ2_16u8u
ippsEncodeHuffFree_BZ2_16u8u

With using these Intel IPP based bzip2 functions, you will be able to follow the building blocks chain defined in bzip2 algorithm and create solution for lossless data compression:

libbzip2 Functions	Corresponding Intel IPP bzip2 functions
BZ2_bzCompressInit	ippsRLEGetSize_BZ2_8u ippsEncodeRLEInit_BZ2_8u ippsEncodeRLEInitAlloc_BZ2_8u ippsBWTFwdGetSize_8u ippsMTFGetSize_8u ippsMTFInit_8u ippsMTFInitAlloc_8u ippsEncodeHuffGetSize_BZ2_16u8u ippsEncodeHuffInit_BZ2_16u8u ippsEncodeHuffInitAlloc_BZ2_16u8u
BZ2_bzCompress	ippsEncodeRLE_BZ2_8u ippsEncodeRLEFlush_BZ2_8u ippsRLEGetInUseTable_8u ippsReduceDictionary_8u_I ippsBWTFwd_8u ippsMTFFwd_8u ippsEncodeZ1Z2_BZ2_8u16u ippsPackHuffContext_BZ2_16u8u ippsEncodeHuff_BZ2_16u8u ippsCRC32_BZ2_8u
BZ2_bzCompressEnd	ippsRLEFree_BZ2_8u ippsHuffFree_BZ2_16u8u ippsMTFFree_8u
BZ2_bzDecompressInit	ippsBWTInvGetSize_8u ippsMTFGetSize_8u ippsMTFInit_8u ippsMTFInitAlloc_8u ippsDecodeHuffGetSize_BZ2_16u8u ippsDecodeHuffInit_BZ2_16u8u ippsDecodeHuffInitAlloc_BZ2_16u8u
BZ2_bzDecompress	ippsDecodeRLE_BZ2_8u ippsBWTInv_8u, ippsMTFInv_8u ippsDecodeZ1Z2_BZ2_16u8u ippsExpandDictionary_8u_I ippsUnackHuffContext_BZ2_16u8u ippsDecodeHuff_BZ2_16u8u ippsCRC32_BZ2_8u
BZ2_bzDecompressEnd	ippsMTFFree_8u ippsHuffFree_BZ2_16u8u
BZ2_bzCompressInit	ippsRLE GetSize_BZ2_8u ippsEncodeRLEInit_BZ2_8u ippsEncodeRLEInitAlloc_BZ2_8u ippsBWTFwdGetSize_8u ippsMTFGetSize_8u, ippsMTFInit_8u ippsMTFInitAlloc_8u ippsEncodeHuffGetSize_BZ2_16u8u ippsEncodeHuffInit_BZ2_16u8u ippsEncodeHuffInitAlloc_BZ2_16u8u
BZ2_bzCompress	ippsEncodeRLE_BZ2_8u ippsEncodeRLEFlush_BZ2_8u ippsRLEGetInUseTable_8u ippsReduceDictionary_8u_I ippsBWTFwd_8u, ippsMTFFwd_8u ippsEncodeZ1Z2_BZ2_8u16u ippsPackHuffContext_BZ2_16u8u ippsEncodeHuff_BZ2_16u8u, ippsCRC32_BZ2_8u
BZ2_bzCompressEnd	ippsRLEFree_BZ2_8u ippsHuffFree_BZ2_16u8u ippsMTFFree_8u
BZ2_bzDecompressInit	ippsBWTInvGetSize_8u ippsMTFGetSize_8u, ippsMTFInit_8u ippsMTFInitAlloc_8u ippsDecodeHuffGetSize_BZ2_16u8u ippsDecodeHuffInit_BZ2_16u8u ippsDecodeHuffInitAlloc_BZ2_16u8u
BZ2_bzDecompress	ippsDecodeRLE_BZ2_8u ippsBWTInv_8u, ippsMTFInv_8u ippsDecodeZ1Z2_BZ2_16u8u ippsExpandDictionary_8u_I ippsUnackHuffContext_BZ2_16u8u ippsDecodeHuff_BZ2_16u8u ippsCRC32_BZ2_8u
BZ2_bzDecompressEnd	ippsMTFFree_8u ippsHuffFree_BZ2_16u8u

Please try these out if you are interested, and let us know your feedback on using these functions.
We also have a plan to integrate these bzip2 features into our data compression sample in our future release, stay tuned.

For Intel IPP 5.2 beta program, please visit http://www.intel.com/software/products/ipp/beta.htm

Hope it helps.
Thanks,
Ying
Intel Corp.

[mb79]

This is a very interesting feature!
Are there any benchmarks available to see what performance acceleration could be reached?

Marco

[Vladimir_Dudnik]

Hi Marco,

our expert has provided the following brief info on that:

Single-threaded ipp_bzip2 performance gain is ~1.3x for both compression and decompression on Calgary Corpus;

Multi-threaded ipp_bzip2 performance gain on Large Corpus is:

for compression is ~2.45x and decompression - ~1.45x on Core2 Duo;

for compression is ~3.2x and decompression - ~1.65x on 2xCore2 Quad;

All performance numbers are obtained at default performance level 9.

Regards,
Vladimir

[mb79]

Thank you Vladimir for this values! I will discuss using IPP with my colleagues now.

Marco

[Vladimir_Dudnik]

Hi Marco,

please note that these numbers are for the latest IPP version, which is 5.3 for now. Currently we are finalizing release IPP 5.3 update 2. The next version, which will be IPP 6.0 beta will became available in several weeks.

Regards,
Vladimir