Hi
what should you do if you are forced to use only C/C++ code on a project and must produce readable code without assembly and other advanced functions
and you need it to perform better
should you choose a different and better compiler?(tested to produce faster code on same source with fastest settings)
and if so why just that?
and use knowledge from assembly to make things faster
I think I want to test clang
If you want to have a same source file for 32/64 bit, C is for it.
gcc and clang have inline assembler support for x64.
clang is a free option, but without CRT and without standard include files and libraries, but have support for MS PDB.
gcc don't have support for MS PDB.
FORCED ?, better to use best compiler/tools for the project.
Also, MinGW64 supports inline assembly.
My take on this is that when we need to use ASM with C or any other high-level language, it is almost always better to use a separate ASM module. The main reason is that, except for very basic things, such as reversing or rotating bits, inline ASM is cumbersome when compared to full fledged ASM. Sure, it will require a call to the module while inline does not, but does it really weight that much in the final computation? Probably not.
Some people believe that C can replace ASM in every instance. It can not, intrinsics are a panacea. When using vector instructions or looking for extreme performance rotines like memory move, better use real ASM.
Every tool has its place, a C compiler can produce very good code in skilled hands but it can also produce garbage when not well written. The main bugbear with C/C++ are its runtime libraries, an experienced C programmer knows how to navigate the available range where an amateur will produce bloated garbage.
Now if you write in MASM, who cares. :rofl:
C/C++ users should know, that masm32 have a msvcrt.lib for msvcrt.dll without VC startup routines.
Now we just have to learn to use them, so we can share simple examples without VC CRT, for example that vcruntime140.dll, what isn't part of Windows OS.
For ANSI console program:
#pragma comment(lib, "msvcrt.lib")
#if _MSC_VER >= 17
# ifdef _WIN64
# pragma comment(linker,"/subsystem:console,5.2")
# else
# pragma comment(linker,"/subsystem:console,5.1")
# endif
#endif
void __cdecl mainCRTStartup(void)
{
__declspec(dllimport) int __cdecl __getmainargs(int*, char***, char***, int, void*);
__declspec(dllimport) void __cdecl exit(int status);
int __cdecl main(int argc, char **argv);
int argc;
char** argv;
char** env;
int si = 0;
__getmainargs(&argc,&argv,&env,0,&si);
exit(main(argc,argv));
}
thanks everyone for your answers,I read clang has both cl and gcc compability mode
its nice to be able to use inline asm also in 64bit mode,also good to use clang because it seems to produce faster code
AW
I want to make a library of this:
working on a 4xsincostanarcsinarccos PROC,which I want to add jumptable and ifs,so you can choose one or several parts of it
also read about C++ argument against inline asm,it can break compilers optimizating on the rest of the code,dont know if its truth in it
Quote from: TimoVJL on August 22, 2019, 09:47:52 PM
C/C++ users should know, that masm32 have a msvcrt.lib for msvcrt.dll without VC startup routines.
Now we just have to learn to use them, so we can share simple examples without VC CRT, for example that vcruntime140.dll, what isn't part of Windows OS.
Problem is msvcrt.dll is not c99 compliant.
You need to fine a way to access ucrtbase.dll which vc has used since VS2015 I believe.
Not sure how Pelles does it?
This link describes one problem:
https://www.oxygenbasic.org/forum/index.php?topic=1947.0
James
There is in fact a good reason for the MASM32/64 versions of MSVCRT, the DLL is part of the system so its functions have been available for a very long time. The VC initialisations are no real use in assembler and are easy enough to code in any case so they are not part of those two import libraries. You can go the path of having to add DLLs for C compilers but surely it defeats the purpose of making stand alone executable files unless you want to create your own DLLs which is a design choice, not a runtime DLL.
Pelles C don't use ucrtbase.dll
ucrt isn't ready yet, it don't have printf, but VC have inline code for it and an additional library for it.
PellesC test with ucrt (https://forum.pellesc.de/index.php?topic=7052.msg26773#msg26773)int main(int argc, char **argv);
void __cdecl mainCRTStartup(void)
{
//_initialize_narrow_environment();
char* pcml[] = {0,_get_narrow_winmain_command_line()};
stdin = __acrt_iob_func(0);
stdout = __acrt_iob_func(1);
stderr = __acrt_iob_func(2);
exit(main(0, pcml));
}
C/C++ programmers, always pray for users to have all redistributables. What runs behind the scenes is a real mess, even experts have trouble understanding it.
But most of this discussion is indeed useless for Masm programmers that produce Masm standalone applications. Fortunately, msvcrt.dll provides everything needed. Things sometimes become tricky when mixing C/C++ and ASM, but the general rule is to use the same libraries/DLLs already used in the high-level part (C/C++).
but there is both cons and pros with DLL's,isnt one problem if you use big arrays,you get an additional memcopy between dll library and main program of those?
Quote from: daydreamer on August 23, 2019, 08:37:04 PM
but there is both cons and pros with DLL's,isnt one problem if you use big arrays,you get an additional memcopy between dll library and main program of those?
I don't think I understand what you mean, but the DLL once loaded is in the same memory space as the main program, no additional memcopy is needed.
Quote from: jcfuller on August 23, 2019, 07:46:33 PMProblem is msvcrt.dll is not c99 compliant.
James,
Excuse my ignorance, but what exactly is the problem there?
QuoteI received an email pointing out a problem with the time functions which I verified.
Output using vs2019 on Win 10 and gcc on Linux:
Code:
Mon Aug 5 14:52:11 2019
Today is Monday, August 05.
The time is 02:52 PM.
ISO week day is 2019W321.
Output from NUWEN and my bc9/TCLib:
Code:
Mon Aug 05 15:18:27 2019
Today is Monday, August 05.
The time is 03:18 PM.
Someone has found one way to crash msvcrt.dll
printf strftime :rofl:
Pelles C
Fri Aug 23 14:57:32 2019
Today is Friday, August 23.
The time is 02:57 PM.
ISO week day is 2019W345.
msvcrt.dll
Fri Aug 23 15:01:31 2019
Today is Friday, August 23.
The time is 03:01 PM.
ucrtbase.dll
Fri Aug 23 15:02:52 2019
Today is Friday, August 23.
The time is 03:02 PM.
ISO week day is 2019W345.
Quote from: jcfuller on August 23, 2019, 07:46:33 PM
Problem is msvcrt.dll is not c99 compliant.
Of course it isn't, as M$ started to conform C99 somewhere 2012, it just took so long.
Now I see the problem - using the C source from James' post at the oxygen forum:
#include <time.h>
#include <stdio.h>
#define SIZE 256
int main (void)
{
char buffer[SIZE];
time_t curtime;
struct tm *loctime;
/* Get the current time. */
curtime = time (NULL);
/* Convert it to local time representation. */
loctime = localtime (&curtime);
/* Print out the date and time in the standard format. */
fputs (asctime (loctime), stdout);
/* Print it out in a nice format. */
strftime (buffer, SIZE, "Today is %A, %B %d.\n", loctime);
fputs (buffer, stdout);
strftime (buffer, SIZE, "The time is %I:%M %p.\n", loctime);
fputs (buffer, stdout);
strftime (buffer, SIZE, "ISO week day is %GW%V%u.\n", loctime);
fputs (buffer, stdout);
return 0;
}
Gcc "forgets" the ISO part, while Pelles C and Visual C print it correctly. Seems more a problem with strftime, though: It returns 0 (success?) for all calls but the last one leaves only a zero byte in the buffer.
The OS msvcrt.dll thing is a good thing for clang, as it don't have own CRT.
So, a small examples can be done with msvcrt.lib from masm32 package, if we forget those pesky include files, other that Windows SDK.
If there is interest in this area, we can share our knowledge of making small programs.
A fare thing ?
I think MSVCRT.DLL continues not recognizing the format specifiers %G and %V. However, the C runtimes after VS 2015 do recognize.
This is for Windows 10.
.model flat, stdcall
size_t typedef ptr
includelib \masm32\lib\msvcrt.lib
time proto C :ptr
strftime proto C :ptr, :size_t, :ptr, :vararg
localtime proto C :ptr
asctime proto C :ptr
puts proto C :ptr
.data
todayIs db "Today is %A, %B %d",0
timeIs db "The time is %I:%M %p.",0
isoMSG db "ISO week day is %GW%V%u.",0
buffer db 256 dup (0)
.code
main proc
LOCAL curtime : qword
LOCAL loctime : ptr
invoke time, 0
mov dword ptr curtime, eax
mov dword ptr curtime+4, edx
invoke localtime, addr curtime
mov loctime, eax
invoke asctime, eax
invoke puts, eax
invoke strftime, addr buffer, 256, offset todayIs, loctime
invoke puts, addr buffer
invoke strftime, addr buffer, 256, offset timeIs, loctime
invoke puts, addr buffer
invoke strftime, addr buffer, 256, offset isoMSG, loctime
invoke puts, addr buffer
ret
main endp
end
Output:
Fri Aug 23 15:34:45 2019
Today is Friday, August 23
The time is 03:34 PM.
(No sign of the ISO message.)
I did find a couple of discussions on MinGW gcc development accessing ucrtbase.dll but I got distracted elsewhere and did not follow up.
All the source for ucrtbase is in the ??\kits\10 folder (I forget the full path now) but all ucrtbase.dll functions I believe rely on exception handling??
James
Quote from: AW on August 24, 2019, 12:41:07 AM
I think MSVCRT.DLL continues not recognizing the format specifiers %G and %V. However, the C runtimes after VS 2015 do recognize.
Sure it stays as a M$ Palmer era :rofl: , it don't change.
Quote from: TimoVJL on August 23, 2019, 10:21:46 PM
Of course it isn't, as M$ started to conform C99 somewhere 2012, it just took so long.
My mistake, i didn't specifically told that msvcrt.dll don't follow that.
But this works, we need to fish ucrt.lib
.model flat, stdcall
size_t typedef ptr
includelib "C:\Program Files (x86)\Windows Kits\10\Lib\10.0.18362.0\ucrt\x86\ucrt.lib"
strftime proto C :ptr, :size_t, :ptr, :vararg
asctime proto C :ptr
puts proto C :ptr
includelib \masm32\lib\msvcrt.lib
localtime proto C :ptr
time proto C :ptr
.data
todayIs db "Today is %A, %B %d",0
timeIs db "The time is %I:%M %p.",0
isoMSG db "ISO week day is %GW%V%u.",0
buffer db 256 dup (0)
.code
main proc
LOCAL curtime : qword
LOCAL loctime : ptr
invoke time, 0
mov dword ptr curtime, eax
mov dword ptr curtime+4, edx
invoke localtime, addr curtime
mov loctime, eax
invoke asctime, eax
invoke puts, eax
invoke strftime, addr buffer, 256, offset todayIs, loctime
invoke puts, addr buffer
invoke strftime, addr buffer, 256, offset timeIs, loctime
invoke puts, addr buffer
invoke strftime, addr buffer, 256, offset isoMSG, loctime
invoke puts, addr buffer
ret
main endp
end
Output:
Fri Aug 23 16:14:41 2019
Today is Friday, August 23
The time is 04:14 PM.
ISO week day is 2019W345.
Reply #9 http://masm32.com/board/index.php?topic=8045.msg88168#msg88168 ucrtbase.def
Quote from: TimoVJL on August 24, 2019, 01:36:37 AM
Reply #9 http://masm32.com/board/index.php?topic=8045.msg88168#msg88168 ucrtbase.def
Yes, ucrtbase.dll is not enough. :sad:
Now ucrtbase.dll is enough. :thumbsup:
.model flat, stdcall
size_t typedef ptr
includelib "C:\Program Files (x86)\Windows Kits\10\Lib\10.0.18362.0\ucrt\x86\ucrt.lib"
;includelib ucrtbase.lib ; works also (supplied by TimoVJL)
strftime proto C :ptr, :size_t, :ptr, :vararg
asctime proto C :ptr
puts proto C :ptr
_localtime32 proto C :ptr
_time64 proto C :ptr
.data
todayIs db "Today is %A, %B %d",0
timeIs db "The time is %I:%M %p.",0
isoMSG db "ISO week day is %GW%V%u.",0
buffer db 256 dup (0)
.code
main proc
LOCAL curtime : qword
LOCAL loctime : ptr
invoke _time64, 0
mov dword ptr curtime, eax
mov dword ptr curtime+4, edx
invoke _localtime32, addr curtime
mov loctime, eax
invoke asctime, eax
invoke puts, eax
invoke strftime, addr buffer, 256, offset todayIs, loctime
invoke puts, addr buffer
invoke strftime, addr buffer, 256, offset timeIs, loctime
invoke puts, addr buffer
invoke strftime, addr buffer, 256, offset isoMSG, loctime
invoke puts, addr buffer
ret
main endp
end
OK, i missed to show somethingtypedef void *FILE;
FILE *__acrt_iob_func(int);
...
FILE stdout = __acrt_iob_func(1);
then fputs() works
stdio.h is a real trouble maker, a compiler specific header, better to avoid it, mostly full of junk, as C standard group allows that.
Yes, it works, (no need for headers :badgrin:):
.model flat, stdcall
size_t typedef ptr
includelib "C:\Program Files (x86)\Windows Kits\10\Lib\10.0.18362.0\ucrt\x86\ucrt.lib"
;includelib ucrtbase.lib ; supplied by TimoVJL
strftime proto C :ptr, :size_t, :ptr, :vararg
asctime proto C :ptr
fputs proto C :ptr, :ptr
_localtime64 proto C :ptr
_time64 proto C :ptr
__acrt_iob_func proto C :BYTE
.data
todayIs db "Today is %A, %B %d",10,0
timeIs db "The time is %I:%M %p.",10,0
isoMSG db "ISO week day is %GW%V%u.",10,0
buffer db 256 dup (0)
localtime equ _localtime64
time equ _time64
.code
main proc
LOCAL curtime : qword
LOCAL loctime : ptr
LOCAL stdout :ptr
invoke time, 0
mov dword ptr curtime, eax
mov dword ptr curtime+4, edx
invoke __acrt_iob_func, 1
mov stdout, eax
invoke localtime, addr curtime
mov loctime, eax
invoke asctime, eax
invoke fputs, eax, stdout
invoke strftime, addr buffer, 256, offset todayIs, loctime
invoke fputs, addr buffer, stdout
invoke strftime, addr buffer, 256, offset timeIs, loctime
invoke fputs, addr buffer, stdout
invoke strftime, addr buffer, 256, offset isoMSG, loctime
invoke fputs, addr buffer, stdout
ret
main endp
end
Using fputs with msvcrt.dll
.model flat, stdcall
FILE struct
_ptr dd ?
_cnt dd ?
_base dd ?
_flag dd ?
_file dd ?
_charbuf dd ?
_bufsiz dd ?
_tmpfname dd ?
FILE ends
includelib \masm32\lib\msvcrt.lib
fputs proto C :ptr, :ptr
externdef C _imp___iob : ptr
stdin equ <_imp___iob>
.data
HelloMsg db "Testing fputs",10,0
.code
main proc
LOCAL stdout : ptr
mov ecx, stdin
add ecx, sizeof FILE
mov stdout, ecx;
mov eax, offset HelloMsg
invoke fputs, eax, stdout
ret
main endp
end
Output:
Testing fputs
James' code built with Cygwin :
$ gcc GetTimeTest.c -o GetTimeTest.exe
$ ./GetTimeTest.exe
Sat Aug 24 12:06:00 2019
Today is Saturday, August 24.
The time is 12:06 PM.
ISO week day is 2019W346.
Quote from: AW on August 23, 2019, 08:46:51 PM
Quote from: daydreamer on August 23, 2019, 08:37:04 PM
but there is both cons and pros with DLL's,isnt one problem if you use big arrays,you get an additional memcopy between dll library and main program of those?
I don't think I understand what you mean, but the DLL once loaded is in the same memory space as the main program, no additional memcopy is needed.
judge for yourself very old unomptimized kinda raycast dll,made with vc++
Magnus,
This one is not a debate, a DLL is mapped into the memory space of the app that calls it, all you have to do to get data from a DLL is to get the pointer to it as it is already in memory. If someone has to copy memory from a DLL to a calling app, they have done it the wrong way. Just pass the pointer.
working example for 64bit??
James
MinGW 64-bit V5.4.0 :
C:\mingw64>gcc GetTimeTest.c -o GetTimeTest.exe
C:\mingw64>GetTimeTest.exe
Sat Aug 24 21:35:29 2019
Today is Saturday, August 24.
The time is 09:35 PM.
ISO week day is W.
Playing with stdio file control structures from msvcrt.dll
#include <tchar.h>
#include <stdio.h>
#define SIZE 256
#undef stdin
#undef stdout
#undef stderr
#pragma comment(lib,"msvcrt.lib")
#ifdef _WIN64
extern char* __imp__iob;
#define stdin (FILE*)&__imp__iob[0]
#define stdout (FILE*)&__imp__iob[sizeof(FILE)]
#define stderr (FILE*)&__imp__iob[2*sizeof(FILE)]
#else
extern char* _imp___iob;
#define stdin (FILE*)&_imp___iob[0]
#define stdout (FILE*)&_imp___iob[sizeof(FILE)]
#define stderr (FILE*)&_imp___iob[2*sizeof(FILE)]
#endif
int mainCRTStartup(void)
{
char buffer[SIZE] = { 0 };
fputs("Was Stalin German, Italian, Russian or Chinese ?:\n", stdout);
while (1)
{
fgets(buffer, SIZE - 1, stdin);
buffer[strcspn(buffer, "\n")] = '\0';
if (_stricmp(buffer, "russian"))
fputs("Wrong, please retry\n", stderr);
else
{
fputs("Correct\n", stdout);
break;
}
}
getchar();
return 0;
}
Calling __p__iob to get STDIN :
.386
.model flat, stdcall
option casemap:none
include \masm32\include\windows.inc
include \masm32\include\kernel32.inc
include \masm32\include\msvcrt.inc
includelib \masm32\lib\kernel32.lib
includelib \masm32\lib\msvcrt.lib
_iobuf STRUCT
_ptr DWORD ?
_cnt DWORD ?
_base DWORD ?
_flag DWORD ?
_file DWORD ?
_charbuf DWORD ?
_bufsiz DWORD ?
_tmpfname DWORD ?
_iobuf ENDS
FILE TYPEDEF _iobuf
.data
msg db 'Message output to STDOUT',0
.data?
stdout dd ?
stdin dd ?
stderr dd ?
.code
start:
call crt___p__iob
mov stdin,eax ; #define stdin (&__iob_func()[0])
mov ecx,SIZEOF(FILE)
add eax,ecx
mov stdout,eax ; #define stdout (&__iob_func()[1])
add eax,ecx
mov stderr,eax ; #define stderr (&__iob_func()[2])
invoke crt_fputs,ADDR msg,stdout
invoke ExitProcess,0
END start
Usual rules of thumb.
- Measure and don't assume.
- Use the best algorithm for the given problem. And you learn this by trying many algorithms.
- Compile without any optimization to debug it troughly. It will be very slow. But it will give your debugger for the debugger to catch the errors you made while drunk at 2am in the morning. :)
- Compile it in release with whatever compiler you have and debug it with whatever debugger you have.
Now you have a product. But if you want to go one extra step.
Then, and only then, you look for a better optimizing compiler. In this case, I would go for Intel C++. It takes ages to do an optimized compile. But it does provide good results for intel CPU (Not sure if the best results for AMD).
Normally it will be a compromise between getting the best running time and getting the best development time.
Quote from: xanatose on March 10, 2020, 07:10:25 AM
Usual rules of thumb.
- Measure and don't assume.
- Use the best algorithm for the given problem. And you learn this by trying many algorithms.
- Compile without any optimization to debug it troughly. It will be very slow. But it will give your debugger for the debugger to catch the errors you made while drunk at 2am in the morning. :)
- Compile it in release with whatever compiler you have and debug it with whatever debugger you have.
Now you have a product. But if you want to go one extra step.
Then, and only then, you look for a better optimizing compiler. In this case, I would go for Intel C++. It takes ages to do an optimized compile. But it does provide good results for intel CPU (Not sure if the best results for AMD).
Normally it will be a compromise between getting the best running time and getting the best development time.
good advices