The LLVM maybe optimize and reorder float operators, it give us different output with GCC/MSVC.
At 2016-02-11 15:57:02,"Pradeep Ramachandran" <[email protected]> wrote: Why is this necessary? Pradeep. On Thu, Feb 11, 2016 at 1:04 PM, <[email protected]> wrote: # HG changeset patch # User Ramya Sriraman <[email protected]> # Date 1455173784 -19800 # Thu Feb 11 12:26:24 2016 +0530 # Node ID b22558e286d95befb2730bcfd8b3613c981ff9d2 # Parent 822782933427f19e76b34f340a3c52fb4bc95424 testbench: Change error range in check_ssim_end4 from 0.0001 to 0.001 diff -r 822782933427 -r b22558e286d9 source/test/pixelharness.cpp --- a/source/test/pixelharness.cpp Tue Feb 09 12:05:42 2016 +0530 +++ b/source/test/pixelharness.cpp Thu Feb 11 12:26:24 2016 +0530 @@ -858,7 +858,7 @@ int width = (rand() % 4) + 1; // range[1-4] float cres = ref(sum0, sum1, width); float vres = checked_float(opt, sum0, sum1, width); - if (fabs(vres - cres) > 0.0001) + if (fabs(vres - cres) > 0.001) return false; reportfail(); _______________________________________________ x265-devel mailing list [email protected] https://mailman.videolan.org/listinfo/x265-devel
_______________________________________________ x265-devel mailing list [email protected] https://mailman.videolan.org/listinfo/x265-devel
