Bugzillas of Interest
Bug 412238 – Investigate use of Sparsehash as a replacement for PLDHash: Consider hash algorithms that are faster or more space-efficient. Currently unassigned.
JPEG
- Bug 411718 – Speed up JPEG decoding by 30% by skipping buffer.
Have the decoder read the input buffer directly instead of allocating
and processing through an intermediate buffer. This is checked in.
- Bug 411379 – Add SSE2 processing for JPEG color, use static instead of dynamic arrays: jdcolor.c optimizations.
Remove the jdcolor processing as Alfred has copied the jdcolor code
into nsJPEGDecoder.cpp - we should add the SSE2 code in there when it
shows up. We're waiting for this to go in.
- Bug 412753 – Speed up JPEG decoding by another 10%: moves jdcolor code into nsJPEGDecoder to avoid Cairo conversion. We're waiting for this to go in.
- Bug 412396 – JPEG decoding/painting 7x slower on trunk v.s. branch on mac: Vlad is working on this. Vlad has a patch but I couldn't get it to apply. Done.
- Bug 256919 – Clean up assembly optimizations of libjpeg:
I think that this was an old version of my IDCT code that was better
than the original Intel code but not as good as my current code. We
should close this. I don't have the ability to close this though.
- Bug 242145 – Optimization of Inverse Discrete Cosine Transform for SSE2: an old attempt to get the SSE2 code in.
- Bug 416157 – Add JPEG SSE2 color processing for ycc_rgb_convert_argb. Waiting for review.
Image
- Bug 406580 – Faster copying of RGB pixel data:
use bswap on jpegs in nsJPEGDecoder.cpp to flip the bytes and add a
0xff. This went in. I have SSE2 code that should do this in half the time.
- Bug 413143 – trunk is broken(sigbus) on SPARC since 20071221. This required aligned reads from the above bug. Where is my SSE2 code?
- Bug 421761 - Use multiplication instead of division for gfxRGBA (gfxColor.h) - entered 3/8/2008
Javascript
- Bug 331545 – Use _InterlockedCompareExchange for js_CompareAndSwap:
there's some interest in this but the Bug 353962 has to be resolved
first. It was checked in with a problem but it didn't break anything
because thin locks are disabled.
- Bug 353962 – Firefox 2.0 often hangs in Intel Mac OS X 10.4.7: catch up on this and respond.
- Bug 414828 – js_strtointeger - use special handling for base 10 .Waiting for review.
PNG
- Bug 428750 – Add PAETH Filter MMX processing back to PNG on Windows
Rounding and Truncation
- Bug 413787 – NSToIntRound, NSToCoordRound code generation. Waiting for review.
nsID::Equals
- Bug 410927 – Consider a reference-equality check in nsID::Equals. WONTFIX
Win64
- Bug 225859 – port NSPR to Windows XP / Server 2003 64bit for AMD64 - done!!!
SIMD
- Bug 413920 – Investigate using SSE4 for CRC32 and String operations
General Performance
- Bug 90198 – Fixed background makes scrolling painfully slow. I guess this is on Windows as I didn't see a problem with Mac OSX.
Bugs
- Bug 426616
– Acid2 incorrect Rendering on FF3b5. Do a binary time search
around early April to find the checkin that caused the problem. Start
with March 1, 2008. The problem exists on Beta 5. Does it exist with
Beta 4?
Updated April 12, 2008.