L00:
Lff:    

flat assembler  version 1.73.08  (4040500 kilobytes memory)
1 passes, 0 bytes.    

L00 Lff does't compile. Why?

But in reality you have no clue about the internal CPU state, and many other things, so any predictions are rendered invalid.

Besides, what is the problem with testing? Why so much resistance to it? Is everyone simply afraid to be proved wrong? By testing we do something called science. You test your assumptions and hypotheses to determine if they are real or imagined. Otherwise you just have a religion. X is better because Furs said it is. Don't ever test it lest we show the folly of such proclamations. Optimisation will be done by fiat!

Haven't read this entire thread, but seeing the good ol' optimization and performance arguments from revolution and furs again.

Noticing a pattern with revolution: you seem to err on the side of 'performance is not predictable until you test it.' And you give valid arguments. Not as familiar with furs' take, but seems to be more of the opposite, that testing is less significant. I agree more with this, but am happy to change my stance.

Question for revolution: There are articles on AMD Manuals such as "Software Optimization Guide for AMD Family 17h Models 30h and Greater Processors". Do these articles have any value in your opinion, if so why or why not?

One quote from article:

Quote:

"Many factors affect instruction execution time. For instance, when a source operand must be loaded from a memory location, the time required to read the operand from system memory adds to the execution time."

Which is your point, but they detail these factors and give specific performance suggestions, such as:

Quote:

"Avoid branches/jumps in the calculation of values."

Performance is a risk assessment, no? It's 100% true that you can't know the exact performance, but you can predict probabilities based on principles, no? I've been doing a lot of Vulkan and SPIR-V lately and I keep having to learn concepts to get frame rates to render faster, like a 30 second clip rendered at 4 seconds per frame instead of it taking 24 hours total.

If I jumped into this thread under the wrong pretext, apologies, feel free to ignore.

Suppose you test and find out binary search is better. You use it. Later you re-test and find out linear search is better, cause stuff changed (?) or you tested differently. Now you rewrite your code?

uselss_cmp_chain        objectively_better_binary       LUTs_are_awesome        computed_goto
2 passes, 1139 bytes.   2 passes, 1137 bytes.           2 passes, 1151 bytes.   3 passes, 1135 bytes.
   1.89 user               1.87 user                       1.88 user               1.87 user
   0.00 system             0.00 system                     0.00 system             0.00 system
0:01.90 elapsed         0:01.87 elapsed                 0:01.88 elapsed         0:01.88 elapsed    

LUTs_are_awesome                                | computed_goto
2 passes, 1151 bytes.   2 passes, 1151 bytes.   | 3 passes, 1135 bytes.   2 passes, 1137 bytes.
  18.86 user              18.80 user            |   18.75 user              18.77 user
   0.00 system             0.34 system          |    0.00 system             0.28 system
0:18.88 elapsed         0:19.16 elapsed         | 0:18.77 elapsed         0:19.07 elapsed
;----------------------------------------------------------------------------------------------
uselss_cmp_chain                                | objectively_better_binary
2 passes, 1139 bytes.   2 passes, 1141 bytes.   | 2 passes, 1137 bytes.   2 passes, 1139 bytes.
  18.89 user              32.81 user            |   18.72 user              19.26 user
   0.57 system             0.01 system          |    0.75 system             0.00 system
0:19.48 elapsed         0:32.85 elapsed         | 0:19.50 elapsed         0:19.28 elapsed    

LUTs_are_awesome                                | computed_goto
2 passes, 1148 bytes.   2 passes, 1148 bytes.   | 3 passes, 1132 bytes.   2 passes, 1134 bytes.
  27.65 user              27.70 user            |   27.63 user              27.68 user
   0.00 system             0.96 system          |    0.01 system             1.02 system
0:27.67 elapsed         0:28.70 elapsed         | 0:27.66 elapsed         0:28.75 elapsed
;----------------------------------------------------------------------------------------------
uselss_cmp_chain                                | objectively_better_binary
2 passes, 1136 bytes.   2 passes, 1138 bytes.   | 2 passes, 1134 bytes.   2 passes, 1136 bytes.
  27.65 user            1:24.48 user            |   32.23 user            2:15.47 user
   0.00 system             0.06 system          |    0.02 system             0.00 system
0:27.67 elapsed         1:24.58 elapsed         | 0:32.28 elapsed         2:15.55 elapsed    

Take a step back and look at your advice here. What did this lead to the poor coder in question to decide for himself: Which code do you write here? Which one do you choose?

You don't see the problem?

How many times do you rewrite this piece of code?

And more importantly, since you have different results everytime, you can't even pick one to satisfy them all!

And for each system we can test it, and then do final assembly, and now we save money and the boss gives us all bonuses. $$$$ If you have never tested code then you won't be aware of all the awesome ways it is useful.

A man is flying in a hot air balloon and realizes he is lost. He reduces height and spots a man down below. He lowers the balloon further and shouts:

"Excuse me, can you help me? Where am I?"

The man below says, "You are in a hot air balloon."

"You must be a programmer," says the balloonist.

"I am," replies the man. "How did you know?"

"Well," says the balloonist, "what you have told me is technically correct, yet totally f&#king worthless."

Good old joke wrote:

A man is flying in a hot air balloon and realizes he is lost. He reduces height and spots a man down below. He lowers the balloon further and shouts:

"Excuse me, can you help me? Where am I?"

The man below says, "You are in a hot air balloon."

"You must be a programmer," says the balloonist.

"I am," replies the man. "How did you know?"

"Well," says the balloonist, "what you have told me is technically correct, yet totally f&#king worthless."

...the boss more likely gives us a big kick out of the company.

Knowledge of what changes might result in significant performance impact is.

A man is flying in a hot air balloon and realizes he is lost. He reduces height and spots a man down below. He lowers the balloon further and shouts:

"Excuse me, can you help me? Where am I?"

The man below says, "You are in a hot air balloon."

"You must be a programmer," says the balloonist.

"I am," replies the man. "How did you know?"

"Well," says the balloonist, "what you have told me is technically correct, yet totally f&#king worthless."

if question=notgoodenough then answer=notgoodenough    

The man below says, "You are in a hot air balloon."

    dec al
    js lff
    dec al
    js lfe
    ...etc
    

As for the original, the smallest for 256 values might be

Did the tests give such contradictory results? Maybe you'll try to analyze the results first before you continue to carry this foolishness of yours. A further test showed that two of the 4 options are objectively better than the others.

...Or you could've just picked qsort from the beginning.

All sorts are slower than some other sort, and all sorts are faster than some other sort. It all depends on your usage.

The data set will dictate which sort works best.

DimonSoft wrote:

...the boss more likely gives us a big kick out of the company.

If you think I should be fired then you'll have to convince my boss with some fantastic argument, because up till now my boss is very pleased to save money. My boss doesn't trust anyone to simply proclaim something superior without proof. And rightly so, we are not a religious organisation.

DimonSoft wrote:

Knowledge of what changes might result in significant performance impact is.

Of course, but the key word there is "might". You won't know until you test it. Otherwise just guessing and merely claiming it is awesome, without any evidence, would get you a very bad review in my company.