mcdonald@uxe.cso.uiuc.edu (09/15/89)
>I am fiddling around with graphic objects in C++ on a PC (EGA mode) >and am frustrated by the time taken to save/restore areas of the >screen bit-map. >I have had a peek at the assembly code doing the pixel copying and it >seems to take around ten instructions per pixel. It is because of poor programming. The EGA has a bizarre but efficient architecture. The ega has a 350x640 pixels by 4 bits mode. That is 112 kbyte of data. So it won't fit, in the obvious way, two pixels to a byte, in one segment. They could have used two segments (a000:0000 to roughly b000:d000) or could have switched portions of map into one segment with an io register bit. But no, they went to separate bitplanes for the four bits, all in one segment. That way they get two whole screens in one segment. Or one 600x800 screen in a segment (on other vendors extended VGA cards). There are extremely efficient ways of copying areas from region to another of video memory. And, to copy to or from ordinary memory, retaining the planar maps, is very fast indeed. So is drawing lines or circles or text in a fixed color. The programmer just has to understand how it works and choose an efficient way to do it. Problem is, and this is evident most clearly in comp.lang.c, is that a lot of programmers learned to program by taking classes in a University somewhere, where they didn't stress enough (at all) the importance of efficiency. And, the graphics textbooks I see all seem to take some abstract model of a graphics device, rather than a real, albeit odd or grubby, one. So people write silly, slow code, not thinking hard enough how to do it fast. IF a programmer understands how the EGA works, it can go blazingly fast. Don't write "modular" or "structured" code: write FAST code, and it will go fast. Doug MCDonald
bobmon@iuvax.cs.indiana.edu (RAMontante) (09/18/89)
mcdonald@uxe.cso.uiuc.edu <110200003@uxe.cso.uiuc.edu> : - -There are extremely efficient ways of copying areas from region -to another of video memory. And, to copy to or from ordinary memory, -retaining the planar maps, is very fast indeed. So is drawing -lines or circles or text in a fixed color. The programmer just has -to understand how it works and choose an efficient way to do it. Okay. Any tips on how to do this?
mcdonald@uxe.cso.uiuc.edu (09/18/89)
>mcdonald@uxe.cso.uiuc.edu <110200003@uxe.cso.uiuc.edu> : >- >-There are extremely efficient ways of copying areas from region >-to another of video memory. And, to copy to or from ordinary memory, >-retaining the planar maps, is very fast indeed. So is drawing >-lines or circles or text in a fixed color. The programmer just has >-to understand how it works and choose an efficient way to do it. >Okay. Any tips on how to do this? To copy from one part of EGA/VGA memory to another you first realize that it is stored (in C) as unsigned char screen[480][640]; that is first the upper left corner, then proceeding across the screen, then down. To copy a region from one place in screen memory to another, just write code to copy the bytes. In C, you might want to use memcpy or memmove. In assembler, use the byte string move instructions. This is done with the control registers of the EGA/VGA in their normal state as set by a bios call to set mode 16 or 18. For example, to copy the top two rows to the 3rd and forth rows: char far *p1; char far *p2; int i,j; p1 = (char far *)0xa000000; p2 = p1 + 1280; for(i = 0; i < 1280; i++)*p2++ = *p1++; (Now for the weird part!!!!! You can replace *p2++ = *p1++ with *p2++ = - *p1++ or *p2++ = *p1++ +12345; and it will still work. Each byte operation in C copies all four bit planes.) To read the color bitplanes to or from ordinary memory, you copy each individual plane one at a time.This requires diddling the IO registers. It is complicated enough that you will need to read a book to see how to set them. Once set, however, the graphics memory behaves just like ordinary memory (no four bytes copied at once). Just be sure to return the registers to normal. Doug MCDonald