When AMD developed x64, instead of creating yet another mode that the processor can be in, they "hijacked" the 16-bit mode. So an x64 CPU can simultaneously juggle (*) 16-bit and 32-bit, or 64-bit and 32-bit, but it cannot juggle 64-bit and 16-bit simultaneously.
Also, its not going to ever change. The design specifically precludes mixing 64-bit and 16-bit because both sets of instructions use the same prefix byte to coerce the following instruction to/from the regular 32-bit version or the "other" version (16-bit or 64-bit.)
And finally, the nail in the coffin, is once in 64-bit mode it cannot get back to 16-bit mode without a power down/reset. This part could be fixed, but that still doesnt do you any good mixing 16-bit and 64-bit.
Lets be honest tho... the only reason that shopkeeper had to make a wedding cake for someone when they didnt want to.. is because the last thing that actually happens when the left plays identity politics... is liberty.
In Spore they dropped their original intent, which was that a creatures capability was conditional on its design, that their capabilities emerged from the simulated physics. That would have been a fun game to play for hundreds, maybe even thousands of hours.
What we actually got was a bullshit opposite of that, of course.
Yes, but it's still going to be ludicrously resource expensive to explore enough of a math space to find a suitable replacement for a given chunk of image.
But thats exactly what modern video codecs do. A frames context is the frames that came before it, and it is this context that is explored with numerous different kinds of views (aka models).. fractal compression just uses a different context (generally IFS.. same technique as produced those old classic "ferns")
Didnt you know? You are acting like an expert... surely you knew? Have I hammered down on you on this very point about knowledge before? You are acting like an expert, but clearly you are not. What happened to honesty? Have I called you a fucking liar before? I'm going to call you one now.
SSE2 and later is very relevant. Not so much regular SSE (1)
SSE essentially doubled the number of registers, and although none of the new registers are fully general purpose thats a huge win. The SIMD portion not so much (its a rare programmer that will use the SIMD extensions of their compiler,) and it didnt help that Intel screwed up their SSE SIMD adding all these "horizontal" operations that defeat the entire advantage of SIMD if you use them.
Those horizontal operations are convenient, but if you are doing high performance SIMD then the data within one of your wide registers is all going to be exactly the same kind and isnt very suitable for any sort of horizontal operations. A registers of high performance SSE SIMD contains X3:X2:X1:X0 (all the same component of a 4 different vectors), not W0:Z0:Y0:X0 (a complete 4-component vector.) The terminology in the SIMD world for these two views of the registers are Structure of Arrays (SoA) and Array of Structures (AoS).
There isnt a single GPU that offers horizontal operations in its instruction set like Intel gave to SSE, because it defeats the purpose and would destroy the performance.
Most x86 programs nowadays rely on "i686" instructions introduced with the Pentium Pro (1995) and Pentium II (1997)
Bullshit. Very few new instructions and not all that useful of ones came out with those CPU's.. and that remained true until SSE2.. not even MMX was useful enough for "most" programs use.
The x86 instruction set minus the SIMD stuff was pretty well fleshed out when the 386 hit, which is a couple generations and a full decade BEFORE the cpu's you are talking about.
Probably the most important single instruction added to the x86/x64 legacy in the past 10 years is POPCNT, which isnt Intels either.
I own the "game." The game in its entirely is switching between models and zooming "in" or "out" to a new level while listening to Alan Watts and "finding" more Alan Watts clips to listen to while roaming around.
I feed it a potentially very small integer, the decoder spits back a large prime number. The decoder does not compute the prime number, it is already computed, and stored in the really big array.
Sigh... if you are storing stuff in the decompressor that is needed to decompress a specific data... then just store the data instead.
You have added layers of complexity here to hide the fact that you dont know what you are talking about... but its not me that you are hiding that fact from... you are hiding that fact from.... you.
Data compression requires a model. The model is used to predict the original data. Corrections are passed along so that the decompressor knows when the model is wrong (like the compressor does)... often these models are built "on-line" as the file is being compressed/decompressed so that the model itself isnt even something thats passed to the decompressor.
There is no reason to think that the mandelbrot would be a good model for predicting anything other than more mandelbrots and related julia's AFAIK.
Good models involve: I've seen something like this before... I'm keeping statistics... the next symbol is X1 with probably P1 or X2 with probability P2, etc...
The compressor could then use a standard entropy encoding technique from here and encode which possibility is really next... range coding or arithmetic coding.. but that actually over-complicates it... instead just a right or wrong indicator is encoded... and THAT indicator is encoding using range/arithmetic because we have a probability for those too.
Indeed it would be quite unusual at this point to learn that this guy was many decades ahead of the professionals in the field, using a strategy he didnt even invent, yet still those professionals arent even close to where he was decades later.
Probably a scam artist, but there have been so many comments here making claims about impossibilities and so forth... at least you know what motivates a scammer... these idiots that dont know what they are talking about here tho... got no idea what motivates them... they just like pretending to be experts because they learned about pigeon holes once.
Information theory stablishes what is really possible
Information theory has very little to do with lossy compression. Lossy compression deals with the specific limits of human perception. The output image only has to be like the input image, not a copy of it. Coherent parts are given an equivalent but easier to encode coherence, and random parts are just labeled by its distribution. Encoding "television snow" is virtually free.
Information is encoded in a way that Shannon would approve... entropy compression... but thats not where lossy compression gets its wins.
The kings of non-lossy entropy compression models right now are two strategies.. one is a prediction by partial matching strategy, the other a context mixing strategy tuned by neural networks. These things can get english language down to just above 1 bit per byte. About 8:1. Thats it. Thats where entropy compression gets you in the realm of the coherence that humans care about.
Lossy then gets you from that ~8:1 range well into the 100:1 and 1000:1 range...
So stop trying to apply Shannon thinking to the problem as some sort of guide for the limits of this. This isnt a Shannon problem. Shannon was concerned with getting the same signal out that came in. These lossy compression schemes have moved way beyond that.
Every model can be used to encode a message to its entropy..... but "its entropy" means "the messages entropy... given the model"... it does not mean "the messages entropy given the best possible model"
The problem with your proposed model is that is has nothing to do with the data. Its simply a trick that you expect will work... but it wont.
I can encode 100% of your random movie files in a single 16-bit LFSR.
Nobody cares about compressing random data.
Thats the fallacy of your argument. People only care about the coherent data.. hollywood movies.. family videos.. there are random elements, but the data is neither arbitrary nor can be said to be random. That pixel next to pixel 148675 is related to it. There is coherence.
It is the coherence that is compressed. The random stuff is just replaced with different random stuff.
..and a tiny fraction of that is interesting, and those that are interesting are so because they arent random noise.
I can encode 100% of the random noise possibilities in only a few bits driving an LFSR... and you wont be able to tell the difference.
Repeat after me: "Everything I learned about the pigeon hole paradox is true, but I didnt actually understand any of it"
This is not a pigeon hole issue because nobody is pretending that these compression technologies do equally well with every kind of data, and even the kind of data (noise for instance, which is traditionally hard to compress) doesn't apply because these compressors are lossy such that random data can be approximated by literally any of a large collection of simple random functions.
You cant tell the difference because if you could then it would have been compressible to begin with. You can only tell when the coherence is amiss, but its the coherence that is "accurately" compressible.. its only the incoherent stuff.. aka noise.. that is hard to encode accurately in only a few bits
We arent even close to the limits of lossy compression. as its just a time/space trade-off.
The problem is that the number of snippets you'd have to store is enormous, and that each snippet itself is going to be a ginormous number (like the bits of the string of bytes in that snippet).
..that problem is that you dont know anything about data compression.
I think on the east coast most land is zoned "residential/agricultural" but they also aren't taxing based on the "zone" that they designated it, but instead on the "assessed value" of the land.
When it doesnt make sense to farm it,... no farm.. different assessment. If its the same assessment either way... well fuck... that is the state creating slaves.. and not useful ones either
Slashdot Mirror
For those unfamiliar with the reasons here...
When AMD developed x64, instead of creating yet another mode that the processor can be in, they "hijacked" the 16-bit mode. So an x64 CPU can simultaneously juggle (*) 16-bit and 32-bit, or 64-bit and 32-bit, but it cannot juggle 64-bit and 16-bit simultaneously.
Also, its not going to ever change. The design specifically precludes mixing 64-bit and 16-bit because both sets of instructions use the same prefix byte to coerce the following instruction to/from the regular 32-bit version or the "other" version (16-bit or 64-bit.)
And finally, the nail in the coffin, is once in 64-bit mode it cannot get back to 16-bit mode without a power down/reset. This part could be fixed, but that still doesnt do you any good mixing 16-bit and 64-bit.
(*) "Thunking" between CPU modes.
Why cant it be both?
Lets be honest tho... the only reason that shopkeeper had to make a wedding cake for someone when they didnt want to.. is because the last thing that actually happens when the left plays identity politics... is liberty.
Remember when the Liberals cared about Liberty?
Mike Rowe seems to be an all-around great thinker and explainer. Wish he were my neighbor or something.
I think its getting attention because Alan Watts has also been getting attention (at least based on the recommendations youtube gives to me.)
I have the "game." The one thing it isnt is an actual game. Get your Alan Watts somewhere else (for instance, youtube as mentioned)
In Spore they dropped their original intent, which was that a creatures capability was conditional on its design, that their capabilities emerged from the simulated physics. That would have been a fun game to play for hundreds, maybe even thousands of hours.
What we actually got was a bullshit opposite of that, of course.
Yes, but it's still going to be ludicrously resource expensive to explore enough of a math space to find a suitable replacement for a given chunk of image.
But thats exactly what modern video codecs do. A frames context is the frames that came before it, and it is this context that is explored with numerous different kinds of views (aka models) .. fractal compression just uses a different context (generally IFS .. same technique as produced those old classic "ferns")
Didnt you know? You are acting like an expert... surely you knew? Have I hammered down on you on this very point about knowledge before? You are acting like an expert, but clearly you are not. What happened to honesty? Have I called you a fucking liar before? I'm going to call you one now.
SSE2 and later is very relevant. Not so much regular SSE (1)
SSE essentially doubled the number of registers, and although none of the new registers are fully general purpose thats a huge win. The SIMD portion not so much (its a rare programmer that will use the SIMD extensions of their compiler,) and it didnt help that Intel screwed up their SSE SIMD adding all these "horizontal" operations that defeat the entire advantage of SIMD if you use them.
Those horizontal operations are convenient, but if you are doing high performance SIMD then the data within one of your wide registers is all going to be exactly the same kind and isnt very suitable for any sort of horizontal operations. A registers of high performance SSE SIMD contains X3:X2:X1:X0 (all the same component of a 4 different vectors), not W0:Z0:Y0:X0 (a complete 4-component vector.) The terminology in the SIMD world for these two views of the registers are Structure of Arrays (SoA) and Array of Structures (AoS).
There isnt a single GPU that offers horizontal operations in its instruction set like Intel gave to SSE, because it defeats the purpose and would destroy the performance.
Most x86 programs nowadays rely on "i686" instructions introduced with the Pentium Pro (1995) and Pentium II (1997)
Bullshit. Very few new instructions and not all that useful of ones came out with those CPU's.. and that remained true until SSE2.. not even MMX was useful enough for "most" programs use.
The x86 instruction set minus the SIMD stuff was pretty well fleshed out when the 386 hit, which is a couple generations and a full decade BEFORE the cpu's you are talking about.
Probably the most important single instruction added to the x86/x64 legacy in the past 10 years is POPCNT, which isnt Intels either.
...that WAS gameplay!
I own the "game." The game in its entirely is switching between models and zooming "in" or "out" to a new level while listening to Alan Watts and "finding" more Alan Watts clips to listen to while roaming around.
What the heck are you talking about?
This stuff isnt Alan Watts philosophy, its just Buddhism. Since you clearly do not know, Alan Watts was the guy who brought Buddhism to the West.
There is no way that they will animate them given the sheer quantity of them. I'm thinking thousands, if not tens of thousands, of models.
I say this as someone who owns the "game." For those looking for an actual game, look elsewhere.
Basically this "game" is a novel way of listening to Alan Watts.
I feed it a potentially very small integer, the decoder spits back a large prime number. The decoder does not compute the prime number, it is already computed, and stored in the really big array.
Sigh... if you are storing stuff in the decompressor that is needed to decompress a specific data... then just store the data instead.
You have added layers of complexity here to hide the fact that you dont know what you are talking about... but its not me that you are hiding that fact from... you are hiding that fact from.... you.
There isnt much software that is 64-bit only... therefore 32-bit emulation is a big deal.
More to the point:
... often these models are built "on-line" as the file is being compressed/decompressed so that the model itself isnt even something thats passed to the decompressor.
... range coding or arithmetic coding.. but that actually over-complicates it... instead just a right or wrong indicator is encoded... and THAT indicator is encoding using range/arithmetic because we have a probability for those too.
Data compression requires a model. The model is used to predict the original data. Corrections are passed along so that the decompressor knows when the model is wrong (like the compressor does)
There is no reason to think that the mandelbrot would be a good model for predicting anything other than more mandelbrots and related julia's AFAIK.
Good models involve: I've seen something like this before... I'm keeping statistics... the next symbol is X1 with probably P1 or X2 with probability P2, etc...
The compressor could then use a standard entropy encoding technique from here and encode which possibility is really next
Indeed it would be quite unusual at this point to learn that this guy was many decades ahead of the professionals in the field, using a strategy he didnt even invent, yet still those professionals arent even close to where he was decades later.
Probably a scam artist, but there have been so many comments here making claims about impossibilities and so forth... at least you know what motivates a scammer... these idiots that dont know what they are talking about here tho... got no idea what motivates them... they just like pretending to be experts because they learned about pigeon holes once.
Its called wavetable synthesis, and it its a compression technology.
Information theory stablishes what is really possible
Information theory has very little to do with lossy compression. Lossy compression deals with the specific limits of human perception. The output image only has to be like the input image, not a copy of it. Coherent parts are given an equivalent but easier to encode coherence, and random parts are just labeled by its distribution. Encoding "television snow" is virtually free.
Information is encoded in a way that Shannon would approve... entropy compression... but thats not where lossy compression gets its wins.
The kings of non-lossy entropy compression models right now are two strategies.. one is a prediction by partial matching strategy, the other a context mixing strategy tuned by neural networks. These things can get english language down to just above 1 bit per byte. About 8:1. Thats it. Thats where entropy compression gets you in the realm of the coherence that humans care about.
Lossy then gets you from that ~8:1 range well into the 100:1 and 1000:1 range...
So stop trying to apply Shannon thinking to the problem as some sort of guide for the limits of this. This isnt a Shannon problem. Shannon was concerned with getting the same signal out that came in. These lossy compression schemes have moved way beyond that.
Nah that doesnt work, at all.
Every model can be used to encode a message to its entropy..... but "its entropy" means "the messages entropy... given the model"... it does not mean "the messages entropy given the best possible model"
The problem with your proposed model is that is has nothing to do with the data. Its simply a trick that you expect will work... but it wont.
given a random movie file
I can encode 100% of your random movie files in a single 16-bit LFSR.
.. family videos.. there are random elements, but the data is neither arbitrary nor can be said to be random. That pixel next to pixel 148675 is related to it. There is coherence.
Nobody cares about compressing random data.
Thats the fallacy of your argument. People only care about the coherent data.. hollywood movies
It is the coherence that is compressed. The random stuff is just replaced with different random stuff.
An 8x8x24-bit rectangle has 10^462 possibilities
I can encode 100% of the random noise possibilities in only a few bits driving an LFSR... and you wont be able to tell the difference.
Repeat after me: "Everything I learned about the pigeon hole paradox is true, but I didnt actually understand any of it"
This is not a pigeon hole issue because nobody is pretending that these compression technologies do equally well with every kind of data, and even the kind of data (noise for instance, which is traditionally hard to compress) doesn't apply because these compressors are lossy such that random data can be approximated by literally any of a large collection of simple random functions.
You cant tell the difference because if you could then it would have been compressible to begin with. You can only tell when the coherence is amiss, but its the coherence that is "accurately" compressible.. its only the incoherent stuff.. aka noise.. that is hard to encode accurately in only a few bits
We arent even close to the limits of lossy compression. as its just a time/space trade-off.
The problem is that the number of snippets you'd have to store is enormous, and that each snippet itself is going to be a ginormous number (like the bits of the string of bytes in that snippet).
People are motivated by incentives.
... no farm .. different assessment. If its the same assessment either way... well fuck... that is the state creating slaves .. and not useful ones either
I think on the east coast most land is zoned "residential/agricultural" but they also aren't taxing based on the "zone" that they designated it, but instead on the "assessed value" of the land.
When it doesnt make sense to farm it,
The disadvantage here is you can't grow any crops on the land while animals are grazing on it.
Sounds like you've got a meat crop to me.
Keeping costs low makes it easier to hide programs that they don't want
sigh....
Is this opposite day for you again?
It is keeping budgets high that allows hiding expenditures, dumbass.