# GLB (wave 5), CONSOLE A/B: the timed scripted match (nav/match60.txt), render thread profile PROF=3 (the clock is read three times per frame; PROF=2 cost 5.6 ms of the 17.4 ms it measured).
# No MSC_VGX_GLSTATS / MSC_VGX_FRAMESTATS here: they add per-draw work (the pair census, the detailed front-end statistics) to a timing run.
# Read in log.txt, every 100 presented frames: '[vgx] prof (... PROF=3 ...): all= stream= clock= net=' (back end ms per frame; stream = the vertex/index upload alone), the '[vgx] per frame:' census (draws,
# streamBytes) and '[vgx] vitaGL: ...' (pool/arena/ring peaks and overruns, programs, patcher, skipped sceGxm sets). Compare windows with the same draws= and streamBytes=.
# C0: the control, every GLB switch off (the glb library with its switches off behaves as the perf library, plus the aligned pool and the counters).
# Needs the glb package (build/np/W5GLB: libvitaGL cdbba42-glb). Run: bash session_np2.sh with ENVFILE=<this file>, or put it as env.txt and start the match script.
MSC_INPUT_SCRIPT=match60.txt
MSC_MAX_FRAMES=12000
MSC_WATCH=1
MSC_WATCH_EVERY=300
MSC_FRAME_TIMING=300
MSC_VITA_MEMSTATS=1
MSC_VITA_RENDER=320x224
MSC_VGX_PROF=3
NP_DATA_DIR=ux0:data/mscharged/np-data
