Brendan G Bohannon

If wanting to look at any of my code, dig around here: github.com/cr88192 for BJX2 stuff look here: github.com/cr88192/bgbtec… etc. This is, at least, for projects I have decided to make the code publicly available.

Idle thought: Maybe people making shows (well, more common in indie shows at this point) should lay off some on the "Hey, our world is a video game!" plot; it is starting to get a little overused/stale, and IMO tends to immediately lessen whatever setting they invoke it in...

Did debate whether to rework the 3R/3RI compare ops to be more consistent, but this didn't seem worth it in cost/benefit (non-zero cost for no functional benefit). As for "Bcc Imm6/Disp6", seems to save around 0.5% off the size of the binary; merit here is debatable...

Well, since I was breaking stuff anyways, also made a few other tweaks that were on my "annoyance list": Moving around a few encodings, and eliminating 32-bit encodings for some other rarely used instructions (demoted to 64-bit encodings), etc. Also added Bcc Imm6+Disp6 ops.

I am on the fence about whether to make a more significant breaking change to my XG3 ISA: Changing the ADD/ADD{S/U}.L instructions from Imm10un to Imm10s. This would allow me to reclaim a reasonable chunk of encoding space with small impact to code density, but breaks stuff, ...

@ProgramMax @lauriewired Can't answer for mainstream CPUs, but for a soft-processor on FPGA, it can make sense to only implement core parts of the ISA in hardware; and effectively implement the rest as firmware. Using such an instruction effectively invokes an invisible trap handler, that fakes the op.

@lauriewired I don't know why, but I had in my mind that it was an elaborate fuse network. On-boot patching makes a lot more sense. Well now I'm curious about the internals of how a CPU handles patchable instructions. I had the wrong mental model. *off to the reading chair*

I feel like this might be a question for @lauriewired Suppose I am running my own local server. I might not want microcode updates that slow down my CPU. The mitigation is a non-issue here. And suppose one CPU is just perfect for my specific case. Market for non-updated CPUs??

@mcnabbd Ironically, if anything IME they are more likely to anthropomorphize themselves, adding sentiment, or hallucinating (personal) experiences they could not have had, etc. But, OTOH, they do allow using a more direct communication style compared with normal human interactions...

i notice that when I chat with the AI, I don't anthropomorphize them at all. i don't imagine I'm talking to a real person, or even a "thing". Like knowing how a magic trick works ruins the trick.

@marcsh @davepl1968 I was originally designing for a customized ISA and originally didn't want to require virtual memory, which created an incentive for smaller stack sizes (mostly ended up with a 128K default), along with an ABI design that allows multiple processes in a single address space, ...

@cr88192 @davepl1968 I havent heard of any C doing scoped malloc as alloca. Does sound awesome, but C is a fertile place to try out great ideas Aside: The stack is just an arena backed by hardware so in my lang (not Clike) I allow replacement of that arena (static&dyn) Still in design

alloca() is fully automatic and MASSIVELY faster since it just bumps the stack pointer. char * buf = alloca(1024); It's not even a runtime call, it's just an intrinsic.

Upsampled to be above the minimum image size, but here is an example of the optimized color palette generated by the SDF encoder... Not sure if there are any obvious patterns here... (Note, luma and proximity sorted during palette generation).

Well, for example, current methods store glyphs as 16x16 pixels, and 16x16 glyphs per page. So, as-is, each page needs 65K as a BMP, and on-average LZ compress to around 30K. Can note that I am using specialized palette optimization algos and no dither (dither makes it worse).

Not sure if anyone knows of a particularly good way to compress SDF images. At present one of my "better" strategies seemingly being to use LZ compressed 256-color BMP images. Some other lossy formats may "look good" but generate worse results on the text-rendering output side.

@icpolicy @davepl1968 Limit depends on target. 4MB or 8MB is typical in Linux on PC class hardware. Windows traditionally used 1MB for stacks. On a custom target of mine, I typically use 128K, but this is more because many programs (even without large local alloc) can break with a 32K or 64K stack.

@davepl1968 Yes but the 8 megabyte limit is a bit of a drawback on systems with gigs of ram, don't you think?

@MuzafferKal_ Yeah, a lot of these sorts of videos cheat in one way or another. This one with a changing character in the corner, and others with bouncing balls to show frame-rates not actually moving at the claimed frame-rates, ... It is annoying sometimes...

The character on the bottom right is changing.

Well, and people more talked about TV shows, a few popular ones in the era: Invader Zim; South Park; Serial Experiments Lain; Gundam; Dragon Ball Z; ... The fail that was "ST: Enterprise". And, games like Half-Life, ... Maybe someone could try to nostalgia-ize this era...

Well, OK, so, say, my school era world wasn't so much about music or tapes; more about computers and tech. Game consoles, people playing Pokemon on Gameboy, ... In HS, people had cellphones (flip phones), with little plushies, like South Park's Kenny, or Gir from Invader Zim...

Hmm... So I am old now (Gen Y); sees a video talking about "Mixtape", felt curious, watches part of an LP. I don't get it / doesn't resonate... I may be old, but not *that* old; and, why do they talk like the Ninja Turtles? (Well, FWIW, my world looked closer to Code Lyoko...).

OK, put a video of it on YouTube, albeit the basic mode (not the 3D glasses mode): youtu.be/beDvAwtf49M ... After a few minutes, almost start to stop noticing that it is monochrome apart from periodic reminders...

May look like a bit much, but added this to the Win32 version, and combining this with 3D anaglyph seems to work reasonably well (3D scene readily visible with 3D glasses despite the harsh dithering effect; and no retinal-rivalry effects...).

Well, another probably pointless experiment: Trying to modify the Doom port to my ISA to have a mode with an aesthetic similar to "Return of the Obra Dinn" (full monochrome). Still a bit lacking though (looks nowhere near as good), and painfully slow ATM...