Further to the two AC posts: you are doing something wrong. As an academic I send / receive latex source all the time and expect it to compile and reproduce the same exact same results. How are you abusing TeX to get these kinds of problems?
No that's not true. A shift instruction has a one cycle latency and 1/2 cycle throughput on the Core2 / Core2-Duo. An add instruction also has a one cycle latency and 1/3 cycle throughput on the Core2-Duo.
The integer multiplier on the Core2-Duo has a 4-cycle throughput and an 8-cycle latency. So in a "simple" case like x*9 = (x<<3)+x the optimisation would take 2 cycles, and the straight mul would take 8. In more complex cases the individual shifts will pipeline for more of a benefit. Only in cases where (t/3)+ceil(lg(t)) >= 8 will the optimisation be as slow as the multiplier for an expansion of t terms (obviously logs in the base 2 as the additions will form a tree). On x86 lacks of registers will kill this optimisation before the cost of the instructions exceeds the multiplier cost.
And yes, I've also benchmarked the code to test it on a Core2-Duo, and my results match Intel's published figures and Agner Fog's data tables so I suggest you recheck your benchmark.
Getting back to the article, the Milepost work isn't really suitable for this type of optimisation anyway. They try and optimise the compile-time by tuning the optimisation flags. For this type of low-level tuning of code the approaches in Program Interpolation, Sketching or PetaBricks would be more appropriate.
Hmmm... that preview thing does its job. I was going to write: "There is quite a big difference in the costs. In the case of cable the revenue from the ads is going to the access provider - so you are paying the same person twice for the content. In the case of internet you are paying for access, and then you are paying the content provider."
But I'm sitting watching an ad-free channel on cable right now and I realise that I'm talking complete crap. The revenue goes to the channel, and they are often different. So for third-party channels you are completely correct, and it's only in the case of channels that the cable company owns (ie Virgin1 or Sky depending on who you are connected to) that the situation is different.
Yes the first. The Dwave guys aren't building quantum computers. Their system lacks entanglement between the qubits, which is essential to running quantum algorithms. They have also been less than forthcoming about the coherence in their system.
Mr. Taucross, what you've just said is one of the most insanely idiotic things I have ever heard. At no point in your rambling, incoherent response were you even close to anything that could be considered a rational thought. Everyone in this room is now dumber for having listened to it. I award you no points, and may God have mercy on your soul.
As for your Apple thing (too lazy to mess with HTML at this point, sorry), the only reason they become popular is because they have advertising. That's all they've got. In every other case, their OS really isn't all that special; it's just ads that convince you they are better.
They have a UNIX system with a full shell that is tightly integrated (through osascript) into a working, polished desktop with a consistent user interface.
Are there any other O/S out there that provide the same?
I'm sorry but that is guff. I'm quite a bit younger than you are (based on your other posts) but in my time in a state comprehensive classes in maths, science and foreign languages were all streamed into sets based on ability.
If you were going to test for primality by sieving then you could take a process that is millions of times slower than the primality test used, and speed it up by a factor of 8.
Instead the test being discussed performs a series of squares and modulo reductions. Each operand is dependent on the previous result - the entire computation is one long dependency chain and so cannot be split onto multiple cores in the way that you describe.
Although having said that, it all flips around again if you look inside the primitive operations that the primality tests uses. So they're using FFT based multiplication steps to do the squaring which obviously can be parallelised quite well..
Your numbers are a little bit off: £347k / 155 = £2239/Mbit/year. Unless your £5100 figure is adding in some other cost as well?
An interesting way to look it is that usage caps in the uk are always GB/month. That 155Mb pipe would transfer 60*60*24*31*155=415152000Mb per month, or 48,330GB/month. The link would cost £39k/month for a price of about £1.24/GB/month. Scary huh? So all of those £20/pm packages with 40GB caps are already loss-leaders...
You're right, I thought it was the network operator rather than the ISP. Although do any of really believe that they operate as entirely separate business units as they are supposed to?
Although it is BT Retail that is referred to in the article as trying to roll the Beeb for loose change, the other references to BT "negotiating" are less clear and could refer to Retail, the network, or even the company as a whole.
BT's packages also have a 40GB soft limit in their FUP - virtually no british home user ADSL ISPs (who use IPstream) offer a truly unlimited service any more, you need to get a business class ADSL account for ã80-100 a month or so.
Slight correction. Be were always quite happy with massive bandwidth usage. I think they do have some kind of soft limit - but it's somewhere north of 400G a month.
That's a really generic argument, so I'm guessing that you are not from the uk. Correct me if I'm wrong.
BT have already squeezed the money out for their upgrade. After the Century-21 roll-out they have enough fibre in place to handle video traffic. After getting to maintain their monopoly for a few years beyond when they should have because DSL didn't fit into the legal view for breaking their monopoly on POTS - they got to rape the entire UK internet industry for bandwidth charges.
They have already collected enough tolls. This is not about who pays for infrastructure upgrades. The backbone is in place, only the exchange endpoints need upgrading and they do not pay for that. Local loop unbundling means that the ISPs pay for that.
The truth is that BT is a monopolist who grew fat collecting tolls. And it wants to find a new place for tolls now that it is not allowed to collect the old ones.
Oh dear. You need to think a bit harder before accusing someone else of spouting bullshit.
Read the final paragraph in this. I trust it is a reliable enough source for you?
I know the width of the widest data register is not related to the size of the address space. But unlike you, I also know that I didn't claim it was - reread what I wrote more carefully. Get an adult to explain the difference between implication and a biconditional, and how informal english differs from logic.
I'm going to restate what I said originally so that it is simple enough for you to understand: 1. The SIMD extensions on the 32-bit x86 instruction set were executed by AMD and Intel's ISAs over multiple cycles. 2. When they transitions to the 64-bit ISA they used to the wider data-path to do more work per cycle. 3. Specifically, they use it in the ALU to handle both single-precision floating-point operations at once. 4. As a result dense numeric code (running over single-precision floats) runs at twice the speed.
A 64-bit machine implies a 64-bit data-path through the processor as well as the 64-bit address space. The numeric calculations that they are referring to are single-precision floating point values accessed through the SIMD extensions. You get twice the throughput on these in 64-bit mode on both AMD and Intel's processors.
In general you're right, but there is actually something real behind that particular claim.
4) The batteries now have way more battery life, which isn't "worsening" the battery situation in my book. Perhaps you're referring to the fact that the battery is not removable? I don't see that as a major issue. How often does a MacBook Pro user replace their battery?
Well, my first one died after 13 months. I'm hoping the second lasts a bit longer. My experience would put me off upgrading to a system where I can't take the battery out.
Shame really, as the 7 or 8hrs on the new systems looks awesome. Especially as they've changed their measurement technique so that is 7 or 8 hours with wifi and realistic usage. I can squeeze 5hrs out of my Santa Rosa Macbook pro with the wifi off working in a shell with vim...
I'm well aware that procedural languages have nothing to do with functional languages. Hence my comment about the separation that occurred with CPL. And I only had one post in this discussion...
It's an interesting point, and because I can see a fair littering of good comments by you all though this discussion I'm intrigued - what did you mean?
K&R derived C from BCPL which was the non-hard-to-compile parts of CPL. The hard parts went on to become the basis of several functional languages.
Actually, if you decide that you can live without and register a completely different address then tell them all of the alternative versions they've missed that you can come up with. Even if it is just a small fee per variation for them to register you are doing your bit to make the whole thing less profitable.
There is a serious flaw in thinking that computers can accurately model macroeconomics, or predict systematic collapses, any better than commonsense and basic logic can.
Why? The rest of your post explains why it was obvious the bubble would burst. But while common sense could tell us that it definitely would burst, it was not so good with the question of when. Many commentators confidently predicted the end of the bubble for years while the economy surged higher on a diet of credit and fairy gold.
So why do you think that we can't derive models more accurate at prediction than common sense?
Amusing. We both have the same amount of programming experience then, and yes, I've coded something in most languages at one point. I also write compilers for a living and spend most days examining how changes to language semantics impact performance.
But hey, if you've finished tossing crap around then you are done. If you feel that you can string a valid point together then continue. In the meantime get the fuck off of my lawn.
Slashdot Mirror
Further to the two AC posts: you are doing something wrong. As an academic I send / receive latex source all the time and expect it to compile and reproduce the same exact same results. How are you abusing TeX to get these kinds of problems?
No that's not true. A shift instruction has a one cycle latency and 1/2 cycle throughput on the Core2 / Core2-Duo. An add instruction also has a one cycle latency and 1/3 cycle throughput on the Core2-Duo.
The integer multiplier on the Core2-Duo has a 4-cycle throughput and an 8-cycle latency. So in a "simple" case like x*9 = (x<<3)+x the optimisation would take 2 cycles, and the straight mul would take 8. In more complex cases the individual shifts will pipeline for more of a benefit. Only in cases where (t/3)+ceil(lg(t)) >= 8 will the optimisation be as slow as the multiplier for an expansion of t terms (obviously logs in the base 2 as the additions will form a tree). On x86 lacks of registers will kill this optimisation before the cost of the instructions exceeds the multiplier cost.
And yes, I've also benchmarked the code to test it on a Core2-Duo, and my results match Intel's published figures and Agner Fog's data tables so I suggest you recheck your benchmark.
Getting back to the article, the Milepost work isn't really suitable for this type of optimisation anyway. They try and optimise the compile-time by tuning the optimisation flags. For this type of low-level tuning of code the approaches in Program Interpolation, Sketching or PetaBricks would be more appropriate.
Hmmm... that preview thing does its job. I was going to write: "There is quite a big difference in the costs. In the case of cable the revenue from the ads is going to the access provider - so you are paying the same person twice for the content. In the case of internet you are paying for access, and then you are paying the content provider."
But I'm sitting watching an ad-free channel on cable right now and I realise that I'm talking complete crap. The revenue goes to the channel, and they are often different. So for third-party channels you are completely correct, and it's only in the case of channels that the cable company owns (ie Virgin1 or Sky depending on who you are connected to) that the situation is different.
Yes the first. The Dwave guys aren't building quantum computers. Their system lacks entanglement between the qubits, which is essential to running quantum algorithms. They have also been less than forthcoming about the coherence in their system.
Watched Pulp Fiction too many times but I can't help but read that in a Christopher Walken voice and expect you to continue:
"when he was shot down over Hanoi he had this laptop with him..."
Mr. Taucross, what you've just said is one of the most insanely idiotic things I have ever heard. At no point in your rambling, incoherent response were you even close to anything that could be considered a rational thought. Everyone in this room is now dumber for having listened to it. I award you no points, and may God have mercy on your soul.
They have a UNIX system with a full shell that is tightly integrated (through osascript) into a working, polished desktop with a consistent user interface.
Are there any other O/S out there that provide the same?
I'm sorry but that is guff. I'm quite a bit younger than you are (based on your other posts) but in my time in a state comprehensive classes in maths, science and foreign languages were all streamed into sets based on ability.
Is that an American style of punctuation? In British English the first (more logical) example is correct.
And others just hate twitter that much.
If you were going to test for primality by sieving then you could take a process that is millions of times slower than the primality test used, and speed it up by a factor of 8.
Instead the test being discussed performs a series of squares and modulo reductions. Each operand is dependent on the previous result - the entire computation is one long dependency chain and so cannot be split onto multiple cores in the way that you describe.
Although having said that, it all flips around again if you look inside the primitive operations that the primality tests uses. So they're using FFT based multiplication steps to do the squaring which obviously can be parallelised quite well..
Because it packs more hype into an n-cube, and fills a 4-dimensional space with marketing.
Come on, that's impressive guys, right?
Try: "where is the colleseum"
Your numbers are a little bit off: £347k / 155 = £2239/Mbit/year. Unless your £5100 figure is adding in some other cost as well?
An interesting way to look it is that usage caps in the uk are always GB/month. That 155Mb pipe would transfer 60*60*24*31*155=415152000Mb per month, or 48,330GB/month. The link would cost £39k/month for a price of about £1.24/GB/month. Scary huh? So all of those £20/pm packages with 40GB caps are already loss-leaders...
You're right, I thought it was the network operator rather than the ISP. Although do any of really believe that they operate as entirely separate business units as they are supposed to?
Although it is BT Retail that is referred to in the article as trying to roll the Beeb for loose change, the other references to BT "negotiating" are less clear and could refer to Retail, the network, or even the company as a whole.
Slight correction. Be were always quite happy with massive bandwidth usage. I think they do have some kind of soft limit - but it's somewhere north of 400G a month.
That's a really generic argument, so I'm guessing that you are not from the uk. Correct me if I'm wrong.
BT have already squeezed the money out for their upgrade. After the Century-21 roll-out they have enough fibre in place to handle video traffic. After getting to maintain their monopoly for a few years beyond when they should have because DSL didn't fit into the legal view for breaking their monopoly on POTS - they got to rape the entire UK internet industry for bandwidth charges.
They have already collected enough tolls. This is not about who pays for infrastructure upgrades. The backbone is in place, only the exchange endpoints need upgrading and they do not pay for that. Local loop unbundling means that the ISPs pay for that.
The truth is that BT is a monopolist who grew fat collecting tolls. And it wants to find a new place for tolls now that it is not allowed to collect the old ones.
Oh dear. You need to think a bit harder before accusing someone else of spouting bullshit.
Read the final paragraph in this. I trust it is a reliable enough source for you?
I know the width of the widest data register is not related to the size of the address space. But unlike you, I also know that I didn't claim it was - reread what I wrote more carefully. Get an adult to explain the difference between implication and a biconditional, and how informal english differs from logic.
I'm going to restate what I said originally so that it is simple enough for you to understand:
1. The SIMD extensions on the 32-bit x86 instruction set were executed by AMD and Intel's ISAs over multiple cycles.
2. When they transitions to the 64-bit ISA they used to the wider data-path to do more work per cycle.
3. Specifically, they use it in the ALU to handle both single-precision floating-point operations at once.
4. As a result dense numeric code (running over single-precision floats) runs at twice the speed.
Think before you reply.
No, you're not quite right.
A 64-bit machine implies a 64-bit data-path through the processor as well as the 64-bit address space. The numeric calculations that they are referring to are single-precision floating point values accessed through the SIMD extensions. You get twice the throughput on these in 64-bit mode on both AMD and Intel's processors.
In general you're right, but there is actually something real behind that particular claim.
Well, my first one died after 13 months. I'm hoping the second lasts a bit longer. My experience would put me off upgrading to a system where I can't take the battery out.
Shame really, as the 7 or 8hrs on the new systems looks awesome. Especially as they've changed their measurement technique so that is 7 or 8 hours with wifi and realistic usage. I can squeeze 5hrs out of my Santa Rosa Macbook pro with the wifi off working in a shell with vim...
Have you replied to the right post?
I'm well aware that procedural languages have nothing to do with functional languages. Hence my comment about the separation that occurred with CPL. And I only had one post in this discussion...
It's an interesting point, and because I can see a fair littering of good comments by you all though this discussion I'm intrigued - what did you mean?
K&R derived C from BCPL which was the non-hard-to-compile parts of CPL. The hard parts went on to become the basis of several functional languages.
Actually, if you decide that you can live without and register a completely different address then tell them all of the alternative versions they've missed that you can come up with. Even if it is just a small fee per variation for them to register you are doing your bit to make the whole thing less profitable.
There is a serious flaw in thinking that computers can accurately model macroeconomics, or predict systematic collapses, any better than commonsense and basic logic can.
Why? The rest of your post explains why it was obvious the bubble would burst. But while common sense could tell us that it definitely would burst, it was not so good with the question of when. Many commentators confidently predicted the end of the bubble for years while the economy surged higher on a diet of credit and fairy gold.
So why do you think that we can't derive models more accurate at prediction than common sense?
Amusing. We both have the same amount of programming experience then, and yes, I've coded something in most languages at one point. I also write compilers for a living and spend most days examining how changes to language semantics impact performance.
But hey, if you've finished tossing crap around then you are done. If you feel that you can string a valid point together then continue. In the meantime get the fuck off of my lawn.