We're talking integer registers here. Everything since the x87 has already supported doubles.
Re:Performance doesn't come directly from 64 bits
on
Is Prescott 64-bit?
·
· Score: 1
x86 has four GPRs, [E]AX through [E]DX (The "E" means 32 bit) and it has four index registers - actually two index registers, and two index offset registers - which can be used with some instructions. Many x86 instructions specify that your result must be stored in a specific register or pair of registers (for 64 bit results of multiply operations for example) and none of those results go into the address index registers.
No, that doesn't matter in practice. The most commonly used integer operations (mov, addition, dereferencing) work with any of the 8 registers. Many instructions that need special registers are so long-running (eg. rep mov) that the extra time to shuffle things into the right registers is negligible.
Finally, consider the worst-case scenario: you really do have 7 numbers that need to be involved in, say, shifting, so they all need to be in ECX at various times. So you need to insert a number of movs or xchgs to shuffle the data around in preparation for each shift. It's still better than spilling to RAM, which you would need to do if you only had 4 registers. Furthermore, inside the P4's trace cache, though I don't know the details, surely all moves get optimized away and the actual code that runs behaves as though it has lots of registers.
In short, x86 performs as though it has 8 GPRs (including the stack pointer), not 4.
Re:Performance doesn't come directly from 64 bits
on
Is Prescott 64-bit?
·
· Score: 1
Clearly 32 bits can only address 4GB of RAM, and for *some* servers more addressing space buys you something. But I'd say they are a very small minority.
Wow, that's an astoundingly ignorant and arrogant statement. I don't even know where to begin.
Re:Performance doesn't come directly from 64 bits
on
Is Prescott 64-bit?
·
· Score: 1
For the short term the Opteron is a pretty impressive chip, but I really don't see how AMD is going to stay on Moore's curve with such a shitty instruction set architecture.
The P4's trace cache has made the ISA largely irrelevant. P4s translate x86 instructions into something more RISC-like internally and then never touch x86 instructions again. It's basically Crusoe in hardware.
With this scheme, x86 is actually a pretty good ISA, because with a trace cache, the only thing about the ISA that matters is code density, and x86's code density is quite good.
Remember, no prediction fails more consistently than one that claims CPUs will not maintain Moore's law.
Re:There's more to it than 64-bit instructions
on
Is Prescott 64-bit?
·
· Score: 1
For one thing, I wonder what its physical external address bus looks like. Can it address more than 4GB of physical memory without paging?
Paging has nothing to do with how much physical memory you can access. Paging is a way to get more virtual memory by caching a large, slow backing store (ie. disk) in physical memory.
The Athlon64 and Opteron can.
So can the Xeon. 32-bit chips can be designed to access any amount of physical ram. It's virtual memory that's typically limited to 32 bits.
With the Prescott and this new extreme edition P4 with 2.5MB cache (I shudder at the yield hit that much cache has per wafer for them)...
Not knowing anything about their fab techniques, I would suspect they would make the cache the way RAM is made: with redundant banks that can be disabled if they are flawed. This would substantially increase the yield despite the extra die area.
Can someone explain that "eat your own dogfood" expression to me? It just doesn't make any sense: what is the analogy supposed to be? Surely people who work at dogfood companies aren't expected to eat the stuff.
Right on. I'm a parent, and I'm appalled that these parents think the whole world should be legally obliged to keep their children safe while they don't lift a finger to raise their own children in a responsible way.
There's a happy medium here, and it's well toward the side of the parent. Society ought to do a reasonable effort not to put undue burdens on parents; for instance, I think it's appropriate that the 6:00 news gives a warning before presenting stories that may be upsetting to children (eg. the death of Mr. Rogers). But having said that, it's my responsibility to keep my son from harm where possible, and teach him to keep himself from harm otherwise.
Prepare not the path for the child; prepare the child for the path.
Sometimes I think these parents ought to be in prison along with (or instead of) the kids.
Insightful? This is about reducing CO2 emmissions using fuel that was previously waste material. How did you come to think it was about mitigating domestic rubbish?
You're right: if it really is a glorified translation, and the author had access to the original code, then it's a copyright violation because it's a derived work.
First, I live in Canada. I know about cold weather. Look at the URL of my home page.
Second, I have a Master's degree in electrical and computer engineering. I know about static electricity.
Third, the POP occurs when the charge you have built up travels to your victim. Travelling charge is a current; it is not static because--guess what--it's travelling. It is static before the POP, but if it stayed static, there'd be no POP at all.
Static electricity, by definition, doesn't move, so it's useless. If you want a motor, you need an electric current. It is possible to turn static electricity into current though, and vice-versa--that's what a capacitor does.
I agree. Even if you became so disgusted with SCO in May that you started to look for a new job, it takes some time to find one. In the mean time, you still need to put food on your plate.
By the time you MD5-hash the line triplets, and then compare the hashes, why don't you just compare the lines in the first place? Seems like it's cheaper and simpler.
If all you want to do is keep the source secret, then a utility to spit out MD5 hashes of each line triple would be sufficient. Then pipe that into "sort | uniq -d" to find duplicate lines. You can even use uniq's "-w" switch to allow you to append line number information to the hashes. Voila, a 1-line shell script that duplicates most of ESR's tool:
Slashdot Mirror
Sorry, why is that ridiculous again?
We're talking integer registers here. Everything since the x87 has already supported doubles.
Finally, consider the worst-case scenario: you really do have 7 numbers that need to be involved in, say, shifting, so they all need to be in ECX at various times. So you need to insert a number of movs or xchgs to shuffle the data around in preparation for each shift. It's still better than spilling to RAM, which you would need to do if you only had 4 registers. Furthermore, inside the P4's trace cache, though I don't know the details, surely all moves get optimized away and the actual code that runs behaves as though it has lots of registers.
In short, x86 performs as though it has 8 GPRs (including the stack pointer), not 4.
With this scheme, x86 is actually a pretty good ISA, because with a trace cache, the only thing about the ISA that matters is code density, and x86's code density is quite good.
Remember, no prediction fails more consistently than one that claims CPUs will not maintain Moore's law.
Can someone explain that "eat your own dogfood" expression to me? It just doesn't make any sense: what is the analogy supposed to be? Surely people who work at dogfood companies aren't expected to eat the stuff.
He invented violence.
There's a happy medium here, and it's well toward the side of the parent. Society ought to do a reasonable effort not to put undue burdens on parents; for instance, I think it's appropriate that the 6:00 news gives a warning before presenting stories that may be upsetting to children (eg. the death of Mr. Rogers). But having said that, it's my responsibility to keep my son from harm where possible, and teach him to keep himself from harm otherwise.
Prepare not the path for the child; prepare the child for the path.
Sometimes I think these parents ought to be in prison along with (or instead of) the kids.
Insightful? This is about reducing CO2 emmissions using fuel that was previously waste material. How did you come to think it was about mitigating domestic rubbish?
You're right: if it really is a glorified translation, and the author had access to the original code, then it's a copyright violation because it's a derived work.
Whatever it is, I can certainly see why it was missed.
This dope gets "5, Interesting" and doesn't even realize Michael didn't say these things? Slashdot's mod system is seriously busted.
Second, I have a Master's degree in electrical and computer engineering. I know about static electricity.
Third, the POP occurs when the charge you have built up travels to your victim. Travelling charge is a current; it is not static because--guess what--it's travelling. It is static before the POP, but if it stayed static, there'd be no POP at all.
Ok, I grant you the balloons, but zapping your friend requires an electric current.
Static electricity, by definition, doesn't move, so it's useless. If you want a motor, you need an electric current. It is possible to turn static electricity into current though, and vice-versa--that's what a capacitor does.
Fair enough. For some reason I thought you were actually referring to the topic of this article. ;-)
Then you don't understand holograms or LCD computer displays.
I agree. Even if you became so disgusted with SCO in May that you started to look for a new job, it takes some time to find one. In the mean time, you still need to put food on your plate.
How can sound travel at 1170km/s through space, but only 340m/s at sea level?
If all you want to do is keep the source secret, then a utility to spit out MD5 hashes of each line triple would be sufficient. Then pipe that into "sort | uniq -d" to find duplicate lines. You can even use uniq's "-w" switch to allow you to append line number information to the hashes. Voila, a 1-line shell script that duplicates most of ESR's tool:
find -name '*.[ch]' -exec codehasher {} \; | sort | uniq -d -w32
Why is ESR's super tool better than this?
Only if "seven metres" is exactly 7.0 metres to begin with. If it's, say, 6.5 metres, then that's 7.1 yars.
Settle down big fella. The quoted number "seven" has only one significant figure. To that standard of accuracy, metres and yards are equivalent.