<html> <head> <base href="https://bugs.freedesktop.org/" /> </head> <body> <div> <a class="bz_bug_link bz_status_NEW " title="NEW - Implement SSBOs in GLSL front-end and i965" href="https://bugs.freedesktop.org/show_bug.cgi?id=89597#c61">Comment # 61</a> on <a class="bz_bug_link bz_status_NEW " title="NEW - Implement SSBOs in GLSL front-end and i965" href="https://bugs.freedesktop.org/show_bug.cgi?id=89597">bug 89597</a> from <a class="email" href="mailto:currojerez@riseup.net" title="Francisco Jerez <currojerez@riseup.net>"> Francisco Jerez</a> <pre>(In reply to Samuel Iglesias from <a href="show_bug.cgi?id=89597#c60">comment #60</a>) > Francisco, I have a question for you related to untyped write messages > source code: > > Today, I have been rebasing our SSBO patch series to master, finding that FS > builder was already pushed upstream and a lot of code was migrated to use > it. I found a crash that is related to one of the patches we picked from you > some time ago that emits untyped write messages. This is the problem: > > When processing nir_intrinsic_store_ssbo in brw_fs_nir.cpp, there is a call > to > surface_access::emit_untyped_write() which ends up internally calling to > emit_header(). > > The function emit_header() (in > src/mesa/drivers/dri/i965/brw_fs_surface_builder.cpp) > copies the sample mask to the header register: > > exec_all(ubld.MOV(component(dst, 7), sample_mask)); > > This MOV fails the following assert in function emit(instruction *inst) at > src/mesa/drivers/dri/i965/brw_fs_builder.h: > > assert(inst->exec_size == dispatch_width() || force_writemask_all); > > According to GDB, those variables have the following values: > > * inst->exec_size = 1 > * dispatch_width() = 8 > * force_writemask_all = 0 > > Removing this assert makes the code to work properly but I wonder if we need > to update our code in some way or if you have already a fix for this. > > If you want to test it, I have published a branch which includes the assert > deletion in the last commit: > > $ git clone -b wip/siglesias/ARB_shader_storage_buffer_object-v2.3 > <a href="https://github.com/Igalia/mesa.git">https://github.com/Igalia/mesa.git</a> > > You can run the following piglit test to reproduce the crash, if you revert > last commit: > > $ bin/arb_shader_storage_buffer_object-layout-std140-write-shader -fbo -auto Ah, yeah, that's because the way channel enable masking is handled has changed since I sent you that branch. Instead of the exec_all() function there's now an exec_all() method in the builder you need to call to get the same effect. I've rebased your branch on top of the latest builder changes (including some changes in the VEC4 builder which is not upstream yet), see the for-samuel branch of my mesa tree: git://people.freedesktop.org/~currojerez/mesa</pre> </div> <hr> You are receiving this mail because: <ul> <li>You are the QA Contact for the bug.</li> </ul> </body> </html>