[NOT] Optimizing compiler

[Everything]

Disclaimer: all of the following applies only to development under Windows!

First of all, I would like to say that I think PB is the most pleasant environment (and language) for writing almost any program.
Apart from some special cases for which I use python (mostly tests, not production), or cases when an assembler is needed, this is always true until the performance of the code comes to the fore and there is an unpleasant moment with Pelle C and its bad optimization.
The compiler’s output code performance is noticeably poorer than GCC, Clang, (Intel is good enough, but kind of specific) and sometimes even MSVC (which optimization is second-rate, at least in non latest cl versions).

In some cases (with minor optimizations), you can make edits to the assembler code and good to go.
In some cases, you can write a part of the code in C++, compile with GCC and import it as a static library.
But sometimes, when you need total optimization of the entire code, you have to completely abandon PB and write everything in C\C++ and compiling in GCC.
It is sad.

As far as I understand, there is no possibility to "replace" the compiler (and not only because of the assembler syntax mismatch).
Is it possible to do something else in the field of code optimization?

[Keya]

Is it possible to do something else in the field of code optimization?

The compiler allows you to output the asm file that it compiles, allowing you to create your own peephole optimiser to modify the asm before it gets compiled

[Mijikai]

You get the whole fasm source that is generated and even before that u can use inline assembly.
Some few things might be slower out of the box (and only noticable in edge cases) but usually the community knows what to do.
So even if you dont know fasm u can pretty much always optimize PB code.

[Everything]

The compiler allows you to output the asm file that it compiles, allowing you to create your own peephole optimiser to modify the asm before it gets compiled

this is exactly what I meant when I wrote

In some cases (with minor optimizations), you can make edits to the assembler code and good to go

When you need some small optimization it's enough, but when everything need to be optimized it's the same that write app in fasm from scratch which makes using high-level language (PB) meaningless in this case moreover, with a large project, this complicates the development so much that it also makes no sense.
Modern optimizing compilers are very smart today, even if you consider yourself an assembler guru in most cases they will be able to generate better code!
For example some of gcc options

[skywalk]

PB in its current form is a rapid prototyping language.
Compiling to ASM instead of C limits the optimization path.
There are hints of LLVM coming and then the playing field will get closer to GCC or CLANG or "insert compiler here".

[Rinzwind]

Example code please?

[infratec]

I also think that it is not a problem of the PB compiler.
Sure, you can speed up everything with asm, but in general PB is very fast.
Most slow stuff is simply not coded in an optimal way.

So yes, a 'slow' example would be nice.

[Everything]

There are hints of LLVM coming

That would be awesome!

There is no 'slow' example
Pelle C output have less performance than OC output (GCC, Clang/LLVM etc) that's fact period
When everything already optimized good compiler is kind of bonus to do a little bit even more.

If people tells you replace your code with asm thats mean:

1. The code you are replacing was unusually inefficient. In most cases improving the algorithm is the most rewarding approach.
2. Your compiler is generating terrible code

Why is that a bad idea?
It's all about development cost. Developing in assembler takes much, much more time than in a higher-level language, and reduces readability and maintainability.

[Mijikai]

There are hints of LLVM coming

That would be awesome!

There is no 'slow' example
Pelle C output have less performance than OC output (GCC, Clang/LLVM etc) that's fact period
When everything already optimized good compiler is kind of bonus to do a little bit even more.

If people tells you replace your code with asm thats mean:

1. The code you are replacing was unusually inefficient. In most cases improving the algorithm is the most rewarding approach.
2. Your compiler is generating terrible code

Why is that a bad idea?
It's all about development cost. Developing in assembler takes much, much more time than in a higher-level language, and reduces readability and maintainability.

Crying about edge cases that every language has and ignoring that most of the issues are solvalble even without asm makes no sense to me.
The whole point of asm (if available) is to do hands on optimization especially if u want to use specific instructions that might not even be available out of the box.
That PB underperforms in its entirety is nonsense.

[Everything]

No one doubts that PB generate perfect code
Although with default PB (PelleC) optimization enabled, this should not be the case.

[mk-soft]

PureBasic is a one-pass compiler, so there may be one PUSH/POP too many when using constants. Value [ (Push) -> Index <- (POP) ]

I had already written that here.
viewtopic.php?f=35&t=70542#p566769

[infratec]

Is it possible to do something else in the field of code optimization?

My first answer: write better code.

But ...

If you are so unsatisfied with all available compilers ... write your own.
So you can optimize everything like you want.

But I'm sure there will be other persons who still find your compiler unoptimized.
Or the compilation time will increase a lot, so that they are unsatisfied with the compilation time.

But what exactly is the target of this post

You don't have 'slow' code which someone of us can optimize.
You only tell us the 'code' is not optimized enough.
Are your programs to slow? Are your programs to large?

Your link to 'perfect code' is nonsense.
Yes, there are additional unneeded instructions. (But in my opinion this is not a bug)
But how often are they used in a large loop, so that they really increase the needed time by a high value?
Yes, it would be nice if this can be removed.
But I don't think that any of my many programs will run noticeably faster after that.

Maybe your post would be better placed in feature requests, since only the PB team can optimize the compiler itself.

[Everything]

better placed in feature requests

I agree.

PB is almost perfect and I want to remove the word "almost".
If you are satisfied no matter what, shame on you

[Tenaja]

If you cannot share exact code, what is the general type of program?

Many of us, myself included, started using pb for its native string handling. However, when passing strings as parameters, full copies of the string are made, then the copy is sent. This is horrendously slow, but it's not advertised in documentation. Unfortunately, for "native PB" code to be faster, you still have to revert to those pointers... And the forums are the only place to learn that at this point.

The point is, of you share your code, or at least the struggling section, you'll likely end up with a better, native, solution... Instead of stressing over asm output or imported libraries.

[idle]

better placed in feature requests

I agree.

PB is almost perfect and I want to remove the word "almost".
If you are satisfied no matter what, shame on you

While there could be improvement in the generated assembly it's not usually the cause of poor performance, it's more to do with the code flow not satisfying branch prediction to take advantage of out of order execution or incurring cache hits all the time. If you're careful you can write very fast code with plain PB even string processing can be done very quickly but it's like any language/compiler you have to learn about it's idiosyncrasies and be mindful that you have to also be a master of herding cats as that's pretty much what optimising code is like theses days.

Here's hoping that PB will evolve