In mips macro assembler, the load16Unaligned(BaseIndex address, RegisterID dest) implementation is not efficient.
Created attachment 269259 [details] Provide a load16Unaligned implementation in mips macro assembler Note: already contributed in qtwekbit http://code.qt.io/cgit/qt/qtwebkit.git/commit/?id=ea22657d17a934b04c8621dc8891a1d4d80510e3
SunSpider's regexp-dna results on my mips board: - without patch ============================================ RESULTS (means and 95% confidence intervals) -------------------------------------------- Total: 353.9ms +/- 0.5% -------------------------------------------- regexp: 353.9ms +/- 0.5% dna: 353.9ms +/- 0.5% - with patch ============================================ RESULTS (means and 95% confidence intervals) -------------------------------------------- Total: 136.5ms +/- 2.6% -------------------------------------------- regexp: 136.5ms +/- 2.6% dna: 136.5ms +/- 2.6%
Guillaume, do you have any feedback on this one ?
(In reply to comment #3) > Guillaume, do you have any feedback on this one ? I couldn't say whether this is the fastest solution, though it might well be. It looks correct to me, and it runs well on my device.
Comment on attachment 269259 [details] Provide a load16Unaligned implementation in mips macro assembler rs=me. I don't understand all the code, but any error here would clearly be isolated to the MIPS platform, Guillaume has given his unofficial r+, it's been used in QtWebKit for almost two years, and no reviewer is working on MIPS.
Comment on attachment 269259 [details] Provide a load16Unaligned implementation in mips macro assembler Works fine here and improves benchmark results. cq+
Comment on attachment 269259 [details] Provide a load16Unaligned implementation in mips macro assembler Clearing flags on attachment: 269259 Committed r203226: <http://trac.webkit.org/changeset/203226>
All reviewed patches have been landed. Closing bug.