<html> <head> <base href="https://bugs.freedesktop.org/" /> </head> <body> <div> <a class="bz_bug_link bz_status_NEW " title="NEW - Unreal Engine demo doesn't run" href="https://bugs.freedesktop.org/show_bug.cgi?id=86469#c4">Comment # 4</a> on <a class="bz_bug_link bz_status_NEW " title="NEW - Unreal Engine demo doesn't run" href="https://bugs.freedesktop.org/show_bug.cgi?id=86469">bug 86469</a> from <a class="email" href="mailto:lemody@gmail.com" title="Tapani Pälli <lemody@gmail.com>"> Tapani Pälli</a> <pre>(In reply to Matt Turner from <a href="show_bug.cgi?id=86469#c3">comment #3</a>) > (In reply to Tapani Pälli from <a href="show_bug.cgi?id=86469#c2">comment #2</a>) > > (In reply to Kenneth Graunke from <a href="show_bug.cgi?id=86469#c1">comment #1</a>) > > > This should actually be pretty easy to fix. Sandybridge has more than 16 > > > MRFs, so we could just adjust the spilling code to use the higher numbered > > > ones which aren't used for anything else. > > > > > > The check also seems a bit heavy handed...it may not actually conflict... > > > > I've checked that Effects works on SNB by simply setting spill_base_mrf 15 > > (as 14 is used). I did not find documentation that would indicate more than > > 16 MRFs available though. Docs indicate there are 16 from which 1 .. 15 can > > be used. > > SNB PRM Vol4 Part2's "Table 5-4. MRF Registers Available in Device Hardware" > says "Number per Thread" - "24 registers" > > This would be an easy experiment for someone with a Sandybridge -- modify > the fs_visitor::spill_reg() function in > src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp Yep, I saw this but bspec states that it might not be safe to use MRFs other than 1 .. 15. It feels a bit risky to go change this just because of one demo.</pre> </div> <hr> You are receiving this mail because: <ul> <li>You are the QA Contact for the bug.</li> </ul> </body> </html>