On Mon, Apr 28, 2014 at 6:07 PM, Volkin, Bradley D <bradley.d.volkin@xxxxxxxxx> wrote: > On Mon, Apr 28, 2014 at 08:53:30AM -0700, Daniel Vetter wrote: >> On Mon, Apr 28, 2014 at 08:22:08AM -0700, bradley.d.volkin@xxxxxxxxx wrote: >> > From: Brad Volkin <bradley.d.volkin@xxxxxxxxx> >> > >> > For clients that submit large batch buffers the command parser has >> > a substantial impact on performance. On my HSW ULT system performance >> > drops as much as ~20% on some tests. Most of the time is spent in the >> > command lookup code. Converting that from the current naive search to >> > a hash table lookup reduces the performance impact by as much as ~10%. >> > >> > The choice of value for I915_CMD_HASH_ORDER allows all commands >> > currently used in the parser tables to hash to their own bucket (except >> > for one collision on the render ring). The tradeoff is that it wastes >> > memory. Because the opcodes for the commands in the tables are not >> > particularly well distributed, reducing the order still leaves many >> > buckets empty. The increased collisions don't seem to have a huge >> > impact on the performance gain, but for now anyhow, the parser trades >> > memory for performance. >> >> For the collisions have you looked into pre-munging the key a bit so that >> we use more bits? A few shifts and xors shouldn't affect perf much really. > > I looked at this briefly but didn't find a substantial improvement. The > basic patch improved things enough that I wanted to just get it out. I > can look into this more, but I'd like to think about implementing the > batch buffer copy portion next. I don't want to optimize this, make people > happy, and then introduce another perf drop from the copy. Better to just > take the full hit now and then continue the improvements. Sound reasonable? Yeah, makes sense. Like I've said just throwing around ideas. -Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
