Re: [OT] On `ad hominem' - OS X users

Hi,

It seems both of you are getting closer to some kind of solution or understanding of why Math::Pari isn't behaving that well on OS-X
and how it could be solved.

It would be a pity though if you couldn't ignore each others insults.

I must agree, that not all comments are nice and that it's not that
easy to stay calm.

About speed gains of assembly vs. C code:
That's a very interesting thread though probably not that perlish.

I think especially with algorithms having very few hot-spot functions and operating with fixed point or multiprecision parameters
assembly code can help. There are sometimes some instructions, which are specifically created for certain algorithms as for example the SAD8 instruction for ARM11 (sum of absolute differenes: subtract 4 pairs of 8 bit values from each other. take the absolute value of each of the four differences and add up their absolute value) has been created for MPEG encoder motion estimation algorithms, which a compiler would never choose by itself. These instructions do accelerate though not always as much as initally expected by the processor designers.

At least for me this thread is interesting. Knowing nothing about OS-X it's nice to learn about the concept of fat libraries and so on.




Sherm Pendley wrote:
Ilya Zakharevich <nospam-abuse@xxxxxxxxx> writes:

Anyway, I'm suspicious of the idea that the asm build is significantly
faster. Compilers have gotten a lot smarter, and CPUs a lot more
complex since the days of counting instruction cycles and optimizing
register use.
Obviously, you do not know how assembler is used in multiprecision
arithmetic... Hint: your processor has 10x number of instructions
than those generated by C compilers...

Why are you being so damned obnoxious? Seriously. It's nothing but one
insult after another with you. What exactly is your problem?

It's a fact that modern compilers *usually* generate faster code than
hand-coded assembler. If this library is an exception to that general
rule, why can't you just say so without being a condescending ass
about it?

BTW: I'm very suspicious of your report of all tests passing... Did
not you just tested 2.1.7?

Yes, because that's the version that it downloaded and patched.

(It's behaviour on OS X is well-understood
[although you still managed to highlight several new facets ;-]; the
discussion was about 3.3.4 - which requires patches I mentioned in my
initial reply of this thread...)

Sorry, but if you want any further help from me, you're going to have
to apologize and lighten up on the hostility.

sherm--

.

Quantcast