Intel's Gallium3D Driver Gains Optimization For Helping With Java OpenGL Performance
In our recent Mesa 19.2 benchmarks of Intel's old and new (Gallium3D) OpenGL Linux drivers one of the rare areas where the new "Iris" driver performed behind the classic driver was with "j2dbench" that stresses the Java OpenGL pipeline. At the time it was unknown why the new driver was performing subpar for this Java graphics test, but now at least there's one optimization so far in addressing that shortcoming.
Intel Iris lead developer Kenneth Graunke didn't know at the time why the Java j2dbench performance was coming up short for the new driver when we presented him with the test results. But now it's looking like he's getting the performance sorted out.
In a commit on Monday to Mesa 19.3-devel he optimized out redundant sampler state binds. Ken explained, "This cuts roughly 85% of the 3DSTATE_SAMPLER_STATE_POINTERS_PS calls in the J2DBench images test. For some reason, the state tracker is calling bind_sampler_state with the same sampler state in a bunch of cases."
Mesa 19.3 is the release due out around December and is also expected to be the release where Intel changes their default open-source Linux OpenGL driver from i965 to Iris for Broadwell graphics and newer.
Intel Iris lead developer Kenneth Graunke didn't know at the time why the Java j2dbench performance was coming up short for the new driver when we presented him with the test results. But now it's looking like he's getting the performance sorted out.
In a commit on Monday to Mesa 19.3-devel he optimized out redundant sampler state binds. Ken explained, "This cuts roughly 85% of the 3DSTATE_SAMPLER_STATE_POINTERS_PS calls in the J2DBench images test. For some reason, the state tracker is calling bind_sampler_state with the same sampler state in a bunch of cases."
Mesa 19.3 is the release due out around December and is also expected to be the release where Intel changes their default open-source Linux OpenGL driver from i965 to Iris for Broadwell graphics and newer.
5 Comments