Direct access to video memory

Read chapter "Structures and types defined by Allegro" for an internal description of the BITMAP structure. There are several ways to get direct access to the image memory of a bitmap, varying in complexity depending on what sort of bitmap you are using.

The simplest approach will only work with memory bitmaps (obtained from create_bitmap(), grabber datafiles, and image files) and sub-bitmaps of memory bitmaps. This uses a table of char pointers, called `line', which is a part of the bitmap structure and contains pointers to the start of each line of the image. For example, a simple memory bitmap putpixel function is: 

   void memory_putpixel(BITMAP *bmp, int x, int y, int color)
   {
      bmp->line[y][x] = color;
   }
For truecolor modes you need to cast the line pointer to the appropriate type, for example: 
   void memory_putpixel_15_or_16_bpp(BITMAP *bmp, int x, int y, int color)
   {
      ((short *)bmp->line[y])[x] = color;
   }

   void memory_putpixel_32(BITMAP *bmp, int x, int y, int color)
   {
      ((long *)bmp->line[y])[x] = color;
   }
If you want to write to the screen as well as to memory bitmaps, you need to use some helper macros, because the video memory may not be part of your normal address space. This simple routine will work for any linear screen, eg. a VESA linear framebuffers: 
   void linear_screen_putpixel(BITMAP *bmp, int x, int y, int color)
   {
      bmp_select(bmp);
      bmp_write8((unsigned long)bmp->line[y]+x, color);
   }
For truecolor modes you should replace the bmp_write8() with bmp_write16(), bmp_write24(), or bmp_write32(), and multiply the x offset by the number of bytes per pixel. There are of course similar functions to read a pixel value from a bitmap, namely bmp_read8(), bmp_read16(), bmp_read24() and bmp_read32().

This still won't work in banked SVGA modes, however, or on platforms like Windows that do special processing inside the bank switching functions. For more flexible access to bitmap memory, you need to call the following routines. They are implemented as inline assembler routines, so they are not as inefficient as they might seem. If the bitmap doesn't require bank switching (ie. it is a memory bitmap, mode 13h screen, etc), these functions just return bmp->line[line].


unsigned long bmp_write_line(BITMAP *bmp, int line);

Selects the line of a bitmap that you are going to draw onto.

Return value: Returns the address of the selected line for writing.

Examples using this: exflame, exlights.
unsigned long bmp_read_line(BITMAP *bmp, int line);

Selects the line of a bitmap that you are going to read from.

Return value: Returns the address of the selected line for reading.

Examples using this: exflame.
void bmp_unwrite_line(BITMAP *bmp);

Releases the bitmap memory after you are finished with it. You only need to call this once at the end of a drawing operation, even if you have called bmp_write_line() or bmp_read_line() several times before it.
Examples using this: exflame, exlights.

More on banked direct memory access

Although SVGA bitmaps are banked, Allegro provides linear access to the memory within each scanline, so you only need to pass a y coordinate to these functions. Various x positions can be obtained by simply adding the x coordinate to the returned address. The return value is an unsigned long rather than a char pointer because the bitmap memory may not be in your data segment, and you need to access it with far pointers. For example, a putpixel using the bank switching functions is: 

   void banked_putpixel(BITMAP *bmp, int x, int y, int color)
   {
      unsigned long address = bmp_write_line(bmp, y);
      bmp_select(bmp);
      bmp_write8(address+x, color);
      bmp_unwrite_line(bmp);
   }
You will notice that Allegro provides separate functions for setting the read and write banks. It is important that you distinguish between these, because on some graphics cards the banks can be set individually, and on others the video memory is read and written at different addresses. Life is never quite as simple as we might wish it to be, though (this is true even when we _aren't_ talking about graphics coding :-) and so of course some cards only provide a single bank. On these the read and write bank functions will behave identically, so you shouldn't assume that you can read from one part of video memory and write to another at the same time. You can call bmp_read_line(), and read whatever you like from that line, and then call bmp_write_line() with the same or a different line number, and write whatever you like to this second line, but you mustn't call bmp_read_line() and bmp_write_line() together and expect to be able to read one line and write the other simultaneously. It would be nice if this was possible, but if you do it, your code won't work on single banked SVGA cards.

And then there's mode-X. If you've never done any mode-X graphics coding, you probably won't understand this, but for those of you who want to know how Allegro sets up the mode-X screen bitmaps, here goes...

The line pointers are still present, and they contain planar addresses, ie. the actual location at which you access the first pixel in the line. These addresses are guaranteed to be quad aligned, so you can just set the write plane, divide your x coordinate by four, and add it to the line pointer. For example, a mode-X putpixel is: 

   void modex_putpixel(BITMAP *b, int x, int y, int color)
   {
      outportw(0x3C4, (0x100<<(x&3))|2);
      bmp_select(bmp);
      bmp_write8((unsigned long)bmp->line[y]+(x>>2), color);
   }

Oh yeah: the DJGPP nearptr hack. Personally I don't like this very much because it disables memory protection and isn't portable to other platforms, but a lot of people swear by it because it can give you direct access to the screen memory via a normal C pointer. Warning: this method will only work with the DJGPP library, when using VGA 13h or a linear framebuffer modes!

In your setup code: 

   #include <sys/nearptr.h>

   unsigned char *screenmemory;
   unsigned long screen_base_addr;

   __djgpp_nearptr_enable();

   __dpmi_get_segment_base_address(screen->seg, &screen_base_addr);

   screenmemory = (unsigned char *)(screen_base_addr + 
                                    screen->line[0] -
                                    __djgpp_base_address);
Then: 
   void nearptr_putpixel(int x, int y, int color)
   {
      screenmemory[x + y*VIRTUAL_W] = color;
   }


Back to contents