Faster Zlib Performance On ARM Thanks To NEON
ARM developer Adenilson Cavalcanti has been working on optimizing the Zlib compression/decompression performance on ARM systems.
By making use of ARM's NEON instructions for SIMD, Adenilson was successful in squeezing better performance out of the ARM chips, among other tuning. With the Chromium Zlib code they were able to make compression 1.36x faster on average or 1.4x for compressing HTML. For decompression speeds they were 1.6x faster with Gzip and 1.8x faster for HTML.
Their end result for Chromium's Zlib (c-zlib) is yielding decompression speeds 1.7~2.0x faster and compression 1.3~1.4x faster. In the process, some of the Intel Zlib performance is also improved. ARM is encouraging users to move over to Chromium's Zlib code.
Those interested in Zlib compression on ARM can see this slide deck from the results presented at this week's Open-Source Summit. Those wanting to experiment with the Chromium Zlib code can find it on Chromium.org.
By making use of ARM's NEON instructions for SIMD, Adenilson was successful in squeezing better performance out of the ARM chips, among other tuning. With the Chromium Zlib code they were able to make compression 1.36x faster on average or 1.4x for compressing HTML. For decompression speeds they were 1.6x faster with Gzip and 1.8x faster for HTML.
Their end result for Chromium's Zlib (c-zlib) is yielding decompression speeds 1.7~2.0x faster and compression 1.3~1.4x faster. In the process, some of the Intel Zlib performance is also improved. ARM is encouraging users to move over to Chromium's Zlib code.
Those interested in Zlib compression on ARM can see this slide deck from the results presented at this week's Open-Source Summit. Those wanting to experiment with the Chromium Zlib code can find it on Chromium.org.
9 Comments