https://bugs.freedesktop.org/show_bug.cgi?id=36295
Summary: Incorrect rendering in SPECviewperf benchmark Product: DRI Version: unspecified Platform: x86-64 (AMD64) OS/Version: Linux (All) Status: NEW Severity: normal Priority: medium Component: DRM/Radeon AssignedTo: dri-devel@lists.freedesktop.org ReportedBy: pelloux@gmail.com
Created an attachment (id=45701) --> (https://bugs.freedesktop.org/attachment.cgi?id=45701) rendering comparison intel vs r600g
Using mesa from GIT + r600g, 7 first tests of SPECviewperf 11 show big rendering errors (see attached image). The same tests run well using an Intel based laptop (with Ubuntu 9.10).
https://bugs.freedesktop.org/show_bug.cgi?id=36295
--- Comment #1 from pelloux@gmail.com 2011-04-16 02:35:29 PDT --- Forgot to mention that tests 3/7 and 4/7 print some Mesa errors when running with ST_DEUG=fallback,mesa :
test3 : Mesa: User error: GL_INVALID_OPERATION in glBegin(fragment program not valid) Mesa: User error: GL_INVALID_OPERATION in glEnd Mesa: User error: GL_INVALID_OPERATION in glProgramStringARB(invalid ARB fragment program option)
test4 : Mesa: User error: GL_INVALID_OPERATION in glProgramStringARB(invalid ARB vertex program option) Mesa: User error: GL_INVALID_OPERATION in glProgramString(bad program) Mesa: User error: GL_INVALID_OPERATION in glProgramStringARB(invalid ARB fragment program option) Mesa: User error: GL_INVALID_OPERATION in glDrawElements(vertex program not valid)
https://bugs.freedesktop.org/show_bug.cgi?id=36295
José Fonseca jfonseca@vmware.com changed:
What |Removed |Added ---------------------------------------------------------------------------- Product|DRI |Mesa Version|unspecified |git Component|DRM/Radeon |Mesa core AssignedTo|dri-devel@lists.freedesktop |mesa-dev@lists.freedesktop. |.org |org CC| |brianp@vmware.com
--- Comment #2 from José Fonseca jfonseca@vmware.com 2011-04-16 02:50:43 PDT --- This is a bug in SPECviewperf11, which relies on NV_fragment_program2 extension, even when it is not advertised.
Me and Brian have a patch that workarounds this by tolerating NV_fragment_program2 's IF/ELSE/ENDIF opcodes.
This is exactly the same stratergy that AMD's fglrx drivers took: not advertise NVIDIA's extension, but silently handle it.
dri-devel@lists.freedesktop.org
-
bugzilla-daemon＠freedesktop.org