TransformationMatrix::inverse should use multiplication instead of division
Reported by
ossy.sze...@gmail.com,
Nov 30 2016
|
||||||
Issue descriptionUserAgent: Mozilla/5.0 (Windows NT 5.1; rv:50.0) Gecko/20100101 Firefox/50.0 Steps to reproduce the problem: The general (non SIMD optimized) path of TransformationMatrix::inverse now divides 16 matrix elements with the determinant. But multiplication with the reciprocal of the determinant would be much faster, at least on ARM(v7). (The SIMD optimizations already uses this trick too.) What is the expected behavior? What went wrong? nothing Did this work before? N/A Chrome version: Channel: n/a OS Version: Flash Version:
,
Nov 30 2016
,
Dec 1 2016
@ossy.szeged-- Could you please help us by providing test case to reproduce the issue and also the chrome browser version in which you are facing the issue. Thanks!
,
Dec 7 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/97e30b35dd6b89ee97f08cf467c80df5d3369fa5 commit 97e30b35dd6b89ee97f08cf467c80df5d3369fa5 Author: ossy.szeged <ossy.szeged@gmail.com> Date: Wed Dec 07 11:56:06 2016 TransformationMatrix::inverse should use multiplication instead of division Microbenchmark: Execute 1 million matrix inversion in a for loop. Result: (measured on Arndale Octa board - Cortex-A15) | | before(ms)| after(ms) | Speedup | | GCC 4.8.2 | 7545 | 2575 | 2.93x | | Clang 4.0.0 (trunk 287685) | 8260 | 3555 | 2.32x | BUG= 669838 Review-Url: https://codereview.chromium.org/2545473002 Cr-Commit-Position: refs/heads/master@{#436920} [modify] https://crrev.com/97e30b35dd6b89ee97f08cf467c80df5d3369fa5/third_party/WebKit/Source/platform/transforms/TransformationMatrix.cpp
,
Dec 26 2016
,
Jan 3 2017
,
Mar 24 2017
|
||||||
►
Sign in to add a comment |
||||||
Comment 1 by ossy.sze...@gmail.com
, Nov 30 2016