Optimize JPG.js receiveAndExtend for 1 bit case. #4894
Conversation
Profiling showed that receiveAndExtend is frequently called with the length of one bit. This happens for example in decodeBaseline. For a single bit, the loop and shift in receive, as well as the shifts in receiveAndExtend are overhead. This shortcut manually calculates the shifts by either returning 1 or -1 from receiveAndExtend by reading the bit and deciding on the return value. While it comes with an overhead for each non-one length, the speedup is at about 10% in the hot parse/decode path.
|
another improvement in the hot path of our documents. For me it gets about 10%. Disclaimer: I do not know the algorithms involved. I just found it interesting that receiveAndExtends returns -1 when the bit is 0. Initially I thought it should be 0. |
|
/botio test |
From: Bot.io (Linux)ReceivedCommand cmd_test from @Snuffleupagus received. Current queue size: 0 Live output at: http://107.21.233.14:8877/84c7968428e4527/output.txt |
From: Bot.io (Windows)ReceivedCommand cmd_test from @Snuffleupagus received. Current queue size: 0 Live output at: http://107.22.172.223:8877/d52226fdd54092b/output.txt |
From: Bot.io (Windows)FailedFull output at http://107.22.172.223:8877/d52226fdd54092b/output.txt Total script time: 21.82 mins
Image differences available at: http://107.22.172.223:8877/d52226fdd54092b/reftest-analyzer.html#web=eq.log |
From: Bot.io (Linux)SuccessFull output at http://107.21.233.14:8877/84c7968428e4527/output.txt Total script time: 26.19 mins
|
|
This commit speeds up |
|
/botio-windows test |
From: Bot.io (Windows)ReceivedCommand cmd_test from @yurydelendik received. Current queue size: 0 Live output at: http://107.22.172.223:8877/76b8f15a540ec51/output.txt |
From: Bot.io (Windows)SuccessFull output at http://107.22.172.223:8877/76b8f15a540ec51/output.txt Total script time: 23.48 mins
|
|
Thank you for the patch |
…xtend Optimize JPG.js receiveAndExtend for 1 bit case.
Profiling showed that receiveAndExtend is frequently called with the length of
one bit. This happens for example in decodeBaseline.
For a single bit, the loop and shift in receive, as well as the shifts in
receiveAndExtend are overhead.
This shortcut manually calculates the shifts by either returning 1 or -1 from
receiveAndExtend by reading the bit and deciding on the return value.
While it comes with an overhead for each non-one length, the speedup is at about
10% in the hot parse/decode path.