<html> <head> <base href="https://bugs.freedesktop.org/"> </head> <body> <div> <a class="bz_bug_link bz_status_NEW " title="NEW - ir_variable has maximum access out of bounds -- but it's not out of bounds" href="https://bugs.freedesktop.org/show_bug.cgi?id=109532#c32">Comment # 32</a> on <a class="bz_bug_link bz_status_NEW " title="NEW - ir_variable has maximum access out of bounds -- but it's not out of bounds" href="https://bugs.freedesktop.org/show_bug.cgi?id=109532">bug 109532</a> from <a class="email" href="mailto:andrey.simiklit.1989@gmail.com" title="andrii simiklit <andrey.simiklit.1989@gmail.com>"> andrii simiklit</a> <pre>(In reply to andrii simiklit from <a href="show_bug.cgi?id=109532#c31">comment #31</a>) > (In reply to Mark Janes from <a href="show_bug.cgi?id=109532#c30">comment #30</a>) > > (In reply to Mark Janes from <a href="show_bug.cgi?id=109532#c28">comment #28</a>) > > > <a href="https://android-review.googlesource.com/c/platform/external/deqp/+/901894">https://android-review.googlesource.com/c/platform/external/deqp/+/901894</a> > > > > Mesa still asserts with this fix. I also tested Andrii's mesa patch with > > the dEQP fix and the test fails. > Do you mean the Chris's dEQP fix here, yes? > But looks like the mentioned Chris's dEQP fix considers some GL limitations > and doesn't affect the expectations of binding points. > > Also the assertion is a separate issue, I created the piglit test for that: > <a href="https://patchwork.freedesktop.org/patch/286287/">https://patchwork.freedesktop.org/patch/286287/</a> > But yes, we unable to fix the test fail without assertion because of crash > :-) > > > > > Since non-mesa drivers have found issues with the original dEQP change, I > > suspect there are still deeper problems with the test. > Possible they have the same issue with binding points mismatch after > optimizations by glsl compiler. > They could try this fix/hack for deqp which is already helped us: > <a href="https://github.com/asimiklit/deqp/commit/">https://github.com/asimiklit/deqp/commit/</a> > 91cff8150944213f6da533e281ee76d95ca00f21 > If it helps them we will know that it is a common issue and it could > expedite this: > <a href="https://github.com/KhronosGroup/OpenGL-API/issues/46">https://github.com/KhronosGroup/OpenGL-API/issues/46</a> So we have an answer from Piers Daniell: "I believe all buffer binding points should be consumed, regardless whether the array elements are used or not. This would be the behavior of least surprise to the developer. I didn't see any language that would indicate that unused elements should not be counted when assigning the element to the buffer binding point." Looks like according to that current mesa implementation is mistaken and dEQP test is right. So my dEQP "fix" should migrate to mesa somehow. Are Ilia and Ian agree it? Is nobody against if I investigate and try to fix it? Seems like the the binding point increment is done somewhere here: <a href="https://gitlab.freedesktop.org/mesa/mesa/blob/master/src/compiler/glsl/link_uniform_blocks.cpp#L279-310">https://gitlab.freedesktop.org/mesa/mesa/blob/master/src/compiler/glsl/link_uniform_blocks.cpp#L279-310</a></pre> </div> <hr> You are receiving this mail because: <ul> <li>You are the assignee for the bug.</li> </ul> </body> </html>