www.delorie.com/djgpp/doc/ug/compiling/runtime.html | search |
This chapter of the user guide discusses what important things to note about the Runtime Environment in DJGPP. Most problems occur because people fail to realize that DJGPP is running in 32-bit mode under DPMI - you do not have as much control over the computer as you would in 16-bit mode.
If you have any comments, suggestions, or questions, feel free to contact the author of this section, Tom St Denis, via email:
The default operand size is the size of the standard unit for arithmetic. The size changes depending on what compiler you use. The only reason that this effects anyone of us, is that the standard unit is also the same size as an int, which is 4-bytes (32-bit). This causes problems when you write programs for 16-bit DOS (where an int is 16-bit) and expect an int to take 2 bytes. For example, using the sizeof operator. Although if you write ANSI-C compatible code this should not affect you. Yet, if you use inline asm or you access hardware ports, you may want to read this. In a 16-bit C Compiler, int's are 16 bits (2 bytes) in size, because it's faster. Same thing for DJGPP, 32-bit ints are faster to use than 16-bit ints. It's not hard to use 16-bit integers in DJGPP. You just use the short keyword. Just make sure that if you are trying to use a 16-bit number use short as this is 16 bits in DJGPP and most popular 16-bit C compilers.
This is about the same as the first topic. Just like int you must note that pointers are 32-bits in size, in DJGPP. In most 16-bit C compilers there are 32-bit pointers, but these consist of a segment and an offset. In DJGPP addresses are linear, meaning that the 32-bit pointer (like char *) is just an offset. Unlike the int scenario, there is no standard way of making a 16-bit pointer. So when using inline assembler, make sure you accept a 32-bit pointer.
This section sums up the previous two, in a condensed format. Unlike 16-bit C compilers, int's are 32-bit, so make sure that you realize this. The safest way to avoid any problems with int's between 32-bit and 16-bit systems, is to use the sizeof operator, instead of assuming the size of the data type. Also do not rely on inline assembly if you are trying to write portable code (between 16 and 32-bit systems). Also just like the int's, pointers are 32-bit too. For example:
void test(int *a, int *b, int amnt) { memcpy(a, b, amnt * 2); }This function copies all the ints in a to b. Can you see the problem in this function? Here look at this:
void test(int *a, int *b, int amnt) { memcpy(a, b, amnt * sizeof(int)); }
Notice the difference? Well the first example will not work correctly in DJGPP. That is just one example of how making assumptions can hurt you.
This section covers the favourite functions in borland that do not work in DJGPP. The first function I wish to talk about is coreleft(). In Borland C++, the coreleft() returns the amount of available heap. This functions is useful because it allows you to see how much ram is available without just running out. However in a DPMI situation where you have dynamic virtual RAM, the heap is not static as you may use virtual ram upon request. You can check how much RAM is available though. Take a look at this example:
#include <stdio.h> #include <alloc.h> void main(void) { printf("Heap size: %lu \n", (unsigned long) coreleft()); }
This displays how much available heap you have. Now you can do something similar in DJGPP too. You use DPMI function 0x0500 to retrieve DPMI memory information, for example:
#include <stdio.h> #include <dpmi.h> int main(void) { __dpmi_free_mem_info dpmi_free_mem; __dpmi_get_free_memory_information(&dpmi_free_mem); printf("Heap: %i\n", dpmi_free_mem.largest_available_free_block_in_bytes); return 0; }
In this example, it displays all the available RAM, including virtual RAM. Next is interrupt support. I will not provide an example here but just to let you know that the getvect()/setvect() functions (dos.h) in Borland C++ do not work the same way in DJGPP. If you use interrupts in your program check out the chapter on interrupts for more information.
There are other Borland specific things that do not work in DJGPP (such as the far keyword, farcalloc, farmalloc, etc...) but this is more a subject for porting 16-bit applications to DJGPP.
Not to long ago, 16-bit compilers ruled the earth. They were considered gods, but soon programmers were feeling tormented, so the gods, grew and became what we call 32-bit compilers. Now they told us stories of great wonders, and pleasures. And that there was. However there were some sacrifices to be made to the gods. One day one of the gods killed his brothers RAW, XMS and VCPI leaving him (DPMI :) ) to rule the 32-bit world.
OK, interesting way to put that. This section covers little things you should know about writing 32-bit programs with DPMI. First, the most import thing that people like to note: DJGPP is 32-bit. Meaning you get more RAM, more space, just plain more. This does however cause confusion. In DJGPP, addresses are always 32 bits, but there are no segments (Actually there are still segments, but they normally are not directly referred to in DJGPP, and not portable to all compilers). This means that the addresses only have an offset to them, this is what we call Linear Addressing. You should note this when porting FAR Data Model programs to DJGPP, because they use segment/offset pair addressing. Now you don't get all the RAM for free though. If you run out of RAM, you can still use more memory (huh?), this is called virtual memory. There is one sacrifice you must make though (drum roll...), performance! Swapping is what they call it when disk space is substituted for RAM. This allows you to load bigger programs, but slower.
Now when you load a program into 32-bit mode, you must have a way to control the program to make sure it doesn't crash on you. So a special program was created to load the programs, it is called a DPMI Server. DPMI stands for Dos Protected Mode Interface, and is covered in the DPMI chapter of the User's Guide. I would just like to let you know that DPMI controls how the program works, accesses ram, and other specialties. Here is a list of some things DPMI provides:
Included in the standard DJGPP C library are all theses functions. DJGPP's runtime environment is based upon DPMI functions. You should only worry about DPMI limitations and functions if your code is not 100% Ansi-C compliant.
This section was provided by Tom St Denis
Email questions or comments to:
webmaster donations bookstore | delorie software privacy |
Copyright © 1998 by DJ Delorie | Updated Sep 1998 |
You can help support this site by visiting the advertisers that sponsor it! (only once each, though)