The MASM Forum

General => The Workshop => Topic started by: clamicun on July 03, 2019, 06:12:36 PM

Title: c++ versus masm
Post by: clamicun on July 03, 2019, 06:12:36 PM
Last week I "translated" one of my Masm programs into c++.
It is not better or worse but the exe file is much larger.
Probably not very well written.

Think I stay with Masm.

For those interested, here is the source compiled with dev-cpp.
My machine: Windows10 64b
Title: Re: c++ versus masm
Post by: TimoVJL on July 03, 2019, 07:00:05 PM
strip off debug-info and size is 918 KB
and without static crt size is 902 KB
Title: Re: c++ versus masm
Post by: clamicun on July 03, 2019, 07:36:44 PM
TimoVJL,
This is the first c++ program I ever wrote.
I haven't the slightest idea what you mean.
Title: Re: c++ versus masm
Post by: TimoVJL on July 03, 2019, 08:21:21 PM
Check this in Dev-C++:
Project -> Project Options ... -> Compiler -> Linker  -> Strip executable (-s) Yes

Ctrl+H opens Project Options.

and then size is something like 920 KB.
Title: Re: c++ versus masm
Post by: clamicun on July 03, 2019, 09:13:55 PM
ok. Thanks
Title: Re: c++ versus masm
Post by: jcfuller on July 04, 2019, 03:31:13 AM
What is the size of your masm version?

James
Title: Re: c++ versus masm
Post by: clamicun on July 04, 2019, 07:06:42 AM
50% smaller
Title: Re: c++ versus masm
Post by: AW on July 04, 2019, 03:21:30 PM
What is more surprising is that the background.bmp alone accounts for 778KB, this is a real miracle :badgrin:

Title: Re: c++ versus masm
Post by: jj2007 on July 04, 2019, 05:53:20 PM
50% smaller
the background.bmp alone accounts for 778KB


Assembly rocks!
Code: [Select]
C++  exe 920 KB - 778KB =  142 KB real code
Masm exe 460 kB - 778KB = -318 KB real code
Title: Re: c++ versus masm
Post by: TimoVJL on July 04, 2019, 06:05:01 PM
.rsrc 909312 bytes 888 KB
.rsrc + .rdata DF000h 913408 bytes 892 KB
Code: [Select]
Dump of file DateiManager.exe

File Type: EXECUTABLE IMAGE

  Summary

        1000 .CRT
        1000 .bss
        1000 .data
        2000 .idata
        1000 .pdata
        1000 .rdata
       DE000 .rsrc
        5000 .text
        1000 .tls
        1000 .xdata
Title: Re: c++ versus masm
Post by: clamicun on July 04, 2019, 06:36:15 PM
Thank you all,
the masm version doesn't use the bmp on the mainwindow.
That's the secret
Title: Re: c++ versus masm
Post by: daydreamer on July 06, 2019, 04:24:09 AM
isnt the right word you "discompiled" your masm code? :biggrin:
(like disassembler,but opposite C++ compiler)
modern C++ compiler has an advantage of not including functions/PROC that are never called and assembly has its advantages of make code small,could it be possible to have conditional assembly do the same thing make assembler code even smaller?
Title: Re: c++ versus masm
Post by: jj2007 on July 06, 2019, 04:50:55 AM
Absolutely, daydreamer! Why don't you test it and post an example?
Title: Re: c++ versus masm
Post by: HSE on July 06, 2019, 05:18:44 AM
...not including functions/PROC that are never called...
If you make assembly programing in a modular way, assemblers don't include functions never called.  :thumbsup:
Title: Re: c++ versus masm
Post by: TimoVJL on July 06, 2019, 10:20:35 AM
uasm supports COMDAT with it's own way.

a smart linker don't include unused code.
Title: Re: c++ versus masm
Post by: hutch-- on July 06, 2019, 12:29:05 PM
Magnus,

The way it works with languages that make object modules that are placed in a library is that any module code that is called is added to the binary by the linker, if it is not called it does not get added. It is called "Linker Resolved Dependencies". If you make the effort to place every algorithm in a separate module, then you have the finest "granularity" and completely exclude unused code.
Title: Re: c++ versus masm
Post by: jcfuller on July 06, 2019, 07:59:39 PM
The tests I did awhile ago with VS 2017 indicated that all functions in #included .cpp source files were compiled and added to the exe. If all code was in a single file only those functions called were added.

Hutch:
  I thought /Gy was applicable when creating object modules with VC++ for a library so I did not have to be as granular with the sources. ???

James
Title: Re: c++ versus masm
Post by: hutch-- on July 06, 2019, 08:36:30 PM
James,

You are talking to the wrong person here, I would not touch C++ with a barge pole. C is OK if written properly and I have seen decent C++ but a lot of it is crap.
Title: Re: c++ versus masm
Post by: TimoVJL on July 06, 2019, 09:37:33 PM
msvc -Gy don't help with C++ classes, as they are like structs. Same problems are with SQLite 3, default functions are included.
Title: Re: c++ versus masm
Post by: daydreamer on July 06, 2019, 11:35:41 PM
The way it works with languages that make object modules that are placed in a library is that any module code that is called is added to the binary by the linker, if it is not called it does not get added. It is called "Linker Resolved Dependencies". If you make the effort to place every algorithm in a separate module, then you have the finest "granularity" and completely exclude unused code.
I am doing it that as much as I can this last year,using IDE and separate source files and a main console test program
I go for similar kind of code in same module


Title: Re: c++ versus masm
Post by: TimoVJL on July 07, 2019, 12:23:13 AM
since ml.exe 12.00 support -Gy , making libs from a one source is possible.
now we have to wait that uasm support it too.
Title: Re: c++ versus masm
Post by: AW on July 07, 2019, 01:09:27 AM
since ml.exe 12.00 support -Gy , making libs from a one source is possible.
now we have to wait that uasm support it too.

UASM probably supports COMDAT because it is mentioned in the JWASM Help file . I never tested, so can't say more.
Title: Re: c++ versus masm
Post by: daydreamer on July 11, 2019, 08:47:43 PM
C++ is maybe more portability,but you can trust masm to use register eax,ebx or any register you want to be what you see is what you get,a C++ compiler may or may not use register keyword and if you use masm to write SIMD packed instructions you get vectors to run fast,otherwise you have to trust compiler vectorize it for you and get performance boost
my suggestion is you let C++ compiler output assembly listing and compare that against your masm assembly listing

C++ changes standard different years,while x86/x64 masm stays the same,except adding support for newer opcodes and support use of bigger registers using same old mnemonics,32 to 64bit,256 to 512 bit
Title: Re: c++ versus masm
Post by: hutch-- on July 11, 2019, 10:37:14 PM
Interestingly enough, ML64 also supports the -Gy switch so when I have some time to play with it, I will see what it does. This much, a library built the old way of one proc for each module, you don't have granularity problems. The -Gy switch may be a convenience but it will not make smaller code that a library written with one procedure for each mosule.
Title: Re: c++ versus masm
Post by: xanatose on August 20, 2019, 03:15:24 AM
One thing I like about assembler is that once you get rid of a bug, it will not come back.
Meanwhile with c++ you just remain with a time bomb until the next compiler change breaks your code.
Or when someone tries to "modernize" perfectly good code. Specially when doing it with automatic tools like clang-tidy. Thank goodness for git.