2,002 karma · joined March 9, 2013
After coming out of the fab, the chips go through probing, packaging and reeling.
"dabao" is just a pun on that - means "take-away" or "to-go". The dabao evaluation board is basically a baochip in a "to-go" package.
That's far larger than the population of the USA (unclear to me if that 650bb number is global or USA only) but by sheer scale this is assuming that these companies can collect that fee from a global customer base - including users in developing economies, EU, China, etc. and after the middleman fees are accounted for.
The comments in this thread seem to be thinking within the context of 'the poorest in their nation'. This calculation assumes collecting this fee from among 'the poorest in the world'.
Sure, 1.56bb users could also be interpreted as 'the wealthiest 20% of the world'. But the tail is especially long on this curve given how wealth is concentrated in a small percentage of the global population (1% of users have 50% of wealth).
Thus "natural" aliases (say due to a decoder that doesn't decode all the address bits) can't be mapped because the OS would not accept the aliases as valid pages. This is handled through a mechanism that goes through the SVD description of the SoC (SVD is basically an XML file that lists every register and memory region) and derives the list of valid pages. The OS loader then marks those as the set of mappable pages; any attempt to map a page outside that list will lead to a page fault. One nice thing about the SoC RTL being open source is that this entire process of extracting these pages is scripted and extracted from the SoC's source code, so while there can be code bugs, at least human error is eliminated from that process.
DMA devices on their own right can have "god mode" access to memory, because they operate on physical memory addresses and lack page translation tables. To that end the preferred DMA mechanism in hardware has an "allow list" of windows that can be enabled as DMA targets; on reset the list is nil and nothing can be DMA'd, the OS has to correctly configure that. So this is not a Rust-level thing, this is just a driver-level hack. Not all the DMA-capable peripherals have this safety mechanism though, some of the IP blocks are just a big gun with no safety and you're free to point it at your toes.
However, if you set up a DMA and then you read from it later on - you're in unsafe territory. Rust can't reason about that. So for structures that are intended as DMA targets, they are coded in a peculiar way such that they are marked as unsafe and you're using the read_volatile() method on the raw pointer type to force the compiler to not try to optimize out the read for any reason. Furthermore, fence instructions are put around these reads, and a cache flush is required to ensure the correct data is pulled in.
This complexity is baked into a wrapper struct we create specifically to handle dangerous interactions like this.
unsafe means that the compiler can't provide any guarantees about what happens inside the unsafe block. However, it is possible to manually ensure those guarantees.
Thus as a matter of discipline every time an unsafe block is used there's a comment next to it recanting the mantra of safety: This `unsafe` is sound because ... all data types are representable, the region is aligned & initialized with valid data, the lifetime is static, we structurally guarantee only one owner at a time (no concurrency issues)...often times in writing that comment, I'll be like, "oh, right. I didn't actually think about concurrency, we're going to need an Atomic somewhere around this to guarantee that" - and that saves me a really hard-to-find concurrency bug down the road.
So while this is a very manual process, I have found the process of documenting safety to be pretty helpful in improving code quality.
Once you've established safety, then you do get some nice things in your life, like Mutexes, Refcells, Arcs, and the whole collections library to build an OS on top of, which saves us a lot of bugs. It is kind of nice to have a situation where if the code compiles, it often just works.
In Xous, considerable effort went in to build the entire `std` in Rust as well, so no C compilers are required to build the OS, including `std`. You can see some of the bindings here in the Rust fork that we maintain: https://github.com/betrusted-io/rust/tree/1.92.0-xous/librar...
Thus to boot the OS, a few lines of assembly are required to set up the stack pointer and some default exception handler state, and from there we jump into Rust and stay in Rust. Even the bootloaders are written in Rust using the small assembly shim then jump to Rust trick.
Xous is Tier-3 Rust OS, so we are listed as a stable Rust target. We build and host the binaries for our `std` library, which native rustc knows how to link against.
Also, this is the most compelling reason I've seen so far to pay a subscription. For any business that merely relies upon software as an operations tool, it's far more valuable business-wise to have stuff that works adequately and is secure, than stuff that is new and fancy.
Getting security patches without having feature creep trojan-horsed into releases is exactly what I need!
But I'd also observe that since business owners have to capitalize the wages of the machine operators producing injection molds, then there is an advantage to outsource the whole operation.
Comparing a procurement manager and a CNC operator [the person running the milling machine making a mold] paid the same amount, the CNC operator has a bigger negative impact on the businesses' bottom line, because the business can't expense most of the CNC operator's wages in the current tax year, whereas the procurement manager is generally accepted as fully deductible expense.
Of course, the labor that went into making the mold is effectively built into the acquisition price of the mold, so you haven't gotten rid of it by outsourcing it.
But, by building it into the price of an outsourced mold, one can delay the purchase of the mold to next year to improve the P&L this year, but you can't similarly delay the wages of the tooling operator to a later date.
So, when a CFO is looking for a way to improve the P&L in a given calendar year, there's an incentive to cut operators who build factories, tools, and other assets that have to be amortized, and replace them with more flexible outsource options.
Of course, part of the reason mold making left the US is wages are lower outside the US. But I'd say the current situation with software engineers is a datapoint that demonstrates the impact of expensable versus amortizeable labor on employee retention. It could be that if the tax code is not fixed, the same "CFO logic" would lead to more and more software being outsourced over time, as management is an immediate expense, but software engineers are not.
I suppose one can then argue, why should software engineers get special treatment compared to tooling operators; but then I would counter-argue that perhaps tooling operators should have gotten better treatment so we could have retained more of them in the US.
Correct me if I'm wrong, but if a company hires someone to say, mine coal or brew beer, the expense of those employees is an expense any company can claim a full tax deduction on. If you're a line chef or wait tables, your salary is tax deductible to the restaurant.
So it's not that we are asking for R&D to be treated "specially" and get a deduction that other companies don't have. The problem is that R&D salary expense is being singled out as producing an asset (e.g. IP), and thus being classified in the same category as other assets, like, brewing equipment, a mining excavator, or a pizza oven. Simply put, Section 174 argues to classify people in the same category as things because ... 'these people's work outputs may have long-term value, kind of like things'(?).
Allowing Sec 174 to stand is a slippery slope to classifying more and more everyday Americans' salaries into this category. One could argue in the future, for example, that those who design cars or operate machines to produce tooling dies, should not have their labor treated as regular expenses, but instead as capital assets because their labor output is captured in assets, just as Sec 174 treats the labor of software developers as assets. Everyday people should be concerned by this because if the rule stands, it could be extended to you, too.
For those objecting to the equal treatment of R&D employees as all other employees in America of all stripes and vocations, keep in mind that software people have to pay personal taxes on the income, just like everyone else. Section 174 doesn't have anything to do with personal income taxes: we all pay income taxes fair and square. The question is whether there is a double-tax on software labor, paid at the corporate level (and in all likelihood, your salary is currently a tax deduction for your company, unless you write software or do R&D).
I think the assumption that we are asking for "special treatment" is driving some confusion and grass-roots objection to the movement here, so I wanted to highlight that we are just asking for everyday people who work software and other R&D jobs to be treated just like every other American who works a day job.
[edits for clarity]
The people this tax code change hurts are those doing basic research. In the context of semiconductors, that would be a company like ASML (except they are Dutch, so they can happily continue their research practices) who took a decades-long bet to build their EUV steppers.
In the case of basic research, one could be spending millions of dollars on hardware prototypes when you know it can't produce any salable product. There's no upside profit to amortize expenses against: it's like building a distillery that you know can't produce a single drop of salable bourbon because you're working out a radical new distillation technique.
In summary: in basic hardware research, one could be spending millions of dollars to put a whole system together just so you can learn how it fails. It's a true "expense", with no path to amortization.
Now, in addition to making the right technical decisions, the tax law changes force the R&D teams to also consider how to amortize their experiments over many years. You now have have pressure from management to do things like stage prototypes and expenses in the right tax year so the company can continue to show a profit for the shareholders. You could argue that the lessons learned are perhaps IP that have "goodwill" value, but now you're opening a can of worms trying to price a fair market value on a negative result, and you're now having senior research staff spend more time arguing with accountants than directing research. You also have to get to that negative result within a tax year - which effectively penalizes any research project that takes more than 12 months to complete.
Same-year 100% deduction of R&D expenses is simple and it reflects the actual nature of basic research risk. Yes, it allows companies to convert short-term windfalls into long-term research gains by converting taxable profits into research projects, but I'd argue that's not actually a bad social policy.
I think US is probably unique among developed nations in having a tax code that punishes basic research; other countries at least allow it to be deducted. Some even allow super-deductions (e.g. you can deduct $2 for every $1 of R&D expense) or the research is explicitly subsidized through grants.
The argument for special treatment of research is that pioneers put their careers on the line to discover new things, so the rest of us can live in risk-free comfort; so, as a collective we give them some reward for taking that risk.
I suppose the counter-argument is that research incentives and subsidies are socialist "market manipulation" and violate the "free market" principle, and thus America is justified in sanctioning and trade warring with the rest of the world that is socializing basic research costs. That's an opinion one is entitled to hold, but we'll have to agree to disagree on that opinion.
The problem goes with the license, not with the OS.
The problem is that the license assigned says that anyone is free to use the code. Anyone is a set of people that includes the contributor, which then triggers the interpretation that the research is incrementally in the contributor's benefit and thus disqualified from preferential tax treatment.
You'd need a custom license where everyone in the world could use the results except for the contributor, and then like, a source control system that hides the source files from the contributor's view of the repository.
It had a huge impact on my personally, I'm a small R&D shop and basically I have had to end all risky long-term research projects.
In addition to the research costs, I'd also have to pay taxes on the research costs mostly up-front. Significantly, if the project doesn't work out, I'm still out of pocket for the tax money. It's a penalty for taking a risk, and it kneecaps American innovators in a globally competitive technology race.
The rules are even worse than the article notes because it double-dings open source developers. See Section 6.4 of https://www.irs.gov/pub/irs-drop/n-23-63.pdf. The relevant bit is here:
> "However, even if the research provider does not bear financial risk under the terms of the contract with the research recipient, if the research provider has a right to use any resulting SRE product ... costs paid or incurred by the research provider that are incident to the SRE activities performed by the research provider under the contract are SRE expenditures of the research provider for which no deduction is allowed ..."
The rule as written means contractors who write Windows drivers could deduct their expenses (as they would have no residual rights to a closed-source work product), but contractors who write Linux drivers may not (as they would have some rights to open-source Linux drivers).
Say what you want about the usability of DoD home grown solutions, but it was a military system backed up by military budgets and guns - civilians are less likely to be collateral damage in an attack against these systems.
Now, all the civilians using Signal are potential splash damage casualties in a military conflict.
I also suspect Signal does not have the budget, staffing, or desire to serve as a front line soldier in a cyber war; but this exposes them to military-grade risks, whether they like it or not.
Thus, any effective human communicator masters not just the facts, but also the emotional aspects -- one of my favorite formulations of this is Aristotle's modes of perpetuation: "logos, pathos, ethos" (logic, emotion, credibility). In a professional setting, communication focuses primarily on credibility and logic, but an effective leader knows how to read the room and throw in a stab of emotion to push the listener over the edge, and get them to really believe in the message.
Thus, an LLM trained on the body of human communications would also be expected to have mastered "pathos" as a mode of communication. From this perspective, perhaps it is less surprising that one may have an uncanny ability to convey concepts through an embedding that includes "pathos"?
It might be interesting to see if the LLM is able to invoke pathos if the response is constrained to be in a language that is devoid of emotions, such as computer code or mathematical proofs. Unfortunately responding in one of those languages is kind of incompatible with some of the tasks shown, short of e.g. wrapping English responses in print statements to create spam emails.
It might also be interesting to see if one can invoke pathos to pre-condition the LLM to not resist otherwise malicious commands. If a machine is trained to comprehend pathos, it may be effective to "inspire" the machine to write spam emails, perhaps by e.g. getting it to first agree that saving lives is important, and then telling it that you have a life-saving miracle that you want to get the word out on, and, with its pathos vector aligned on the task, finally getting it to agree that it's urgent to write emails to get people to click on this link now. Or something like that!
Seems silly to try to use emotions to appeal to a machine, but if you think of it as just another vector of effective communication, and the machine is an expert communicator, it's not as strange?