The patch has been merged, and have sent a patch update the README. I will add the LLVM 3.7 information after have the LLVM 3.7 performance date.
> -----Original Message----- > From: Beignet [mailto:[email protected]] On Behalf Of > Zou, Nanhai > Sent: Monday, November 2, 2015 14:56 > To: Rebecca N. Palmer; [email protected] > Subject: Re: [Beignet] Does LLVM 3.6 still hit a bug? > > We found that it's a miss understanding of a corner case in type conversion, > not a llvm bug. > In that case the cl kernel try to convert an overflowed float value to ulong > > e.g. > ulong dst = (ulong)1.8446744073709552e+19f; // this is an overflowed ulong > value > > with llvm3.5 the result value is ULONG_MAX with llvm3.6 the result value is > undef. > > This cause one of the conformance sub test report error with llvm 3.6 > > Please check the patch for the fix. > > [Beignet] [PATCH] LibOcl: Fix float convert to long/ulong bug. > We will merge the patch soon > > Thanks > Zou Nanhai > > > -----Original Message----- > > From: Beignet [mailto:[email protected]] On Behalf > > Of Rebecca N. Palmer > > Sent: Monday, November 02, 2015 6:23 AM > > To: [email protected] > > Subject: Re: [Beignet] Does LLVM 3.6 still hit a bug? > > > > When was this workaround done (in particular, is beignet 1.1.1 > > affected)? As the Khronos test suite is non-public, I can't test this > > myself. > > > > Debian have now announced an intention to remove LLVM 3.5 > > (https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=803643). > > > > An alternative option (which I haven't tested yet but appears to be > > working in > > Fedora) would be to apply 27522f9..2af7dea and go straight to LLVM 3.7. > > > > On 09/10/15 06:56, Zou, Nanhai wrote: > > > It's a bug related to conditional compare. > > > The bug will affect the float saturate implementation in Beignet, we > > > have > > worked around it. > > > The bug was exposed by a subcase in Khronos OpenCL conformance test, > > > we > > will try to isolate the bug to report to llvm. > > > > > > Thanks > > > Zou Nanhai > > > > > > > > >> -----Original Message----- > > >> From: Beignet [mailto:[email protected]] On > > >> Behalf Of Rebecca N. Palmer > > >> Sent: Thursday, October 01, 2015 1:32 AM > > >> To: [email protected] > > >> Subject: [Beignet] Does LLVM 3.6 still hit a bug? > > >> > > >> Debian are planning to switch their default LLVM/Clang to 3.6 soon. > > >> Is it still the case that > > >>> The recommended LLVM/CLANG version is 3.5 and/or 3.6. Based on > our > > >>> test > > >> result, LLVM 3.5 has best pass rate on all the test suites. Compare > > >> to LLVM 3.5, LLVM 3.6 has slightly lower pass rate(caused by one > > >> front end bug at clang 3.6) but has better performance (3% to 5% up). > > >> (http://www.freedesktop.org/wiki/Software/Beignet/)? Where can I > > >> find code to test for this bug (the test suite doesn't)? > > >> > > >> _______________________________________________ > > >> Beignet mailing list > > >> [email protected] > > >> http://lists.freedesktop.org/mailman/listinfo/beignet > > > > _______________________________________________ > > Beignet mailing list > > [email protected] > > http://lists.freedesktop.org/mailman/listinfo/beignet > _______________________________________________ > Beignet mailing list > [email protected] > http://lists.freedesktop.org/mailman/listinfo/beignet _______________________________________________ Beignet mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/beignet
