@steffo I need to measure register spill overhead, and the method we've come up with is to make the stack uncached. This should theoretically be the worst-case where other running threads steal your cache lines... I guess?

@steffo I tried to run SANE_DEBUG_AIRSCAN=true scanimage -L and I saw something interesting:WSDD: HTTP POST http://192.168.1.88:80/WebServices/Device: 200 OKWSDD: HTTP POST http://192.168.1.88:80/WebServices/Device: OKWSDD: "Brother MFC-L3740CDW series": IPv6 address expected but not yet discoveredWSDD: "Brother MFC-L3740CDW series": IPv6 address expected but not yet discoveredWSDD: "Brother MFC-L3740CDW series": IPv6 address expected but not yet discovered...I guess that sane-airscan saw the IPv4 printer and somehow tried to contact it in IPv6, and it just failed

@steffo Yes.I got an Obsidian page open called "Tessellation Shader". I'm reading 3 blogs, 2 specifications and 1 hour long video.It's both a live-blogging thing and a cry for help.

@steffo yess! You just do a reverse next and you can inspect the state of the system before that instruction was executed. It's the holy grail of debugging.Some emulators might have these commands built-in, but for x86-64 it seems to magically records the changes to the registers and the memory, and when you undo an instruction it enters replay mode.It should be just like "ohh, my program is trying to fetch code ad 0x0000, press undo, ah here was the jump"

@steffo I helped a lot but I couldn't have made it without @gfxstrand and other contributors from the mesa team!NAK is really wonderful.