Accessing the VGA Display Memory
The standard VGA hardware contains up to 256K of onboard display memory. While it would seem logical that this memory would be directly available to the processor, this is not the case. The host CPU accesses the display memory through a window of up to 128K located in the high memory area. (Note that many SVGA chipsets provide an alternate method of accessing video memory directly, called a Linear Frame Buffer.) Thus in order to be able to access display memory you must deal with registers that control the mapping into host address space. To further complicate things, the VGA hardware provides support for memory models similar to that used by the monochrome, CGA, EGA, and MCGA adapters. In addition, due to the way the VGA handles 16 color modes, additional hardware is included that can speed access immensely. Also, hardware is present that allows the programer to rapidly copy data from one area of display memory to another. While it is quite complicated to understand, learning to utilize the VGA's hardware at a low level can vastly improve performance. Many game programmers utilize the BIOS mode 13h, simply because it offers the simplest memory model and doesn't require having to deal with the VGA's registers to draw pixels. However, this same decision limits them from being able to use the infamous X modes, or higher resolution modes.
Detecting the Amount of Display Memory on the Adapter
Most VGA cards in existence have 256K on board; however there is the possibility that some VGA boards have less. To actually determine further if the card has 256K one must actually write to display memory and read back values. If RAM is not present in a location, then the value read back will not equal the value written. It is wise to utilize multiple values when doing this, as the undefined result may equal the value written. Also, the card may alias addresses, causing say the same 64K of RAM to appear 4 times in the 256K address space, thus it is wise to change an address and see if the change is reflected anywhere else in display memory. In addition, the card may buffer one location of video memory in the chipset, making it appear that there is RAM at an address where there is none present, so you may have to read or write to a second location to clear the buffer. Not that if the Extended Memory field is not set to 1, the adapter appears to only have 64K onboard, thus this bit should be set to 1 before attempting to determine the memory size.
Mapping of Display Memory into CPU Address Space
The first element that defines this mapping is whether or not the VGA decodes accesses from the CPU. This is controlled by the RAM Enable field. If display memory decoding is disabled, then the VGA hardware ignores writes to its address space. The address range that the VGA hardware decodes is based upon the Memory Map Select field. The following table shows the address ranges in absolute 32-bit form decoded for each value of this field:
- 00 -- A0000h-BFFFFh -- 128K
- 01 -- A0000h-AFFFFh -- 64K
- 10 -- B0000h-B7FFFh -- 32K
- 11 -- B8000h-BFFFFh -- 32K
Host Address to Display Address Translation
The most complicated part of accessing display memory involves the translation between a host address and a display memory address. Internally, the VGA has a 64K 32-bit memory locations. These are divided into four 64K bit planes. Because the VGA was designed for 8 and 16 bit bus systems, and due to the way the Intel chips handle memory accesses, it is impossible for the host CPU to access the bit planes directly, instead relying on I/O registers to make part of the memory accessible. The most straightforward display translation is where a host access translates directly to a display memory address. What part of the particular 32-bit memory location is dependent on certain registers and is discussed in more detail in Manipulating Display Memory below. The VGA has three modes for addressing, Chain 4, Odd/Even mode, and normal mode:
- Chain 4: This mode is used for MCGA emulation in the 320x200 256-color mode. The address is mapped to memory MOD 4 (shifted right 2 places.)
Manipulating Display Memory
The VGA hardware contains hardware that can perform bit manipulation on data and allow the host to operate on all four display planes in a single operation. These features are fairly straightforward, yet complicated enough that most VGA programmers choose to ignore them. This is unfortunate, as properly utilization of these registers is crucial to programming the VGA's 16 color modes. Also, knowledge of this functionality can in many cases enhance performance in other modes including text and 256 color modes. In addition to normal read and write operations the VGA hardware provides enhanced operations such as the ability to perform rapid comparisons, to write to multiple planes simultaneously, and to rapidly move data from one area of display memory to another, faster logical operations (AND/OR/XOR) as well as bit rotation and masking.
Reading from Display Memory
The VGA hardware has two read modes, selected by the Read Mode field. The first is a straightforward read of one or more consecutive bytes (depending on whether a byte, word or dword operation is used) from one bit plane. The value of the Read Map Select field is the page that will be read from. The second read mode returns the result of a comparison of the display memory and the Color Compare field and masked by the Color Don't Care field. This mode which can be used to rapidly perform up to 32 pixel comparisons in one operation in the planar video modes, helpful for the implementation of fast flood-fill routines. A read from display memory also loads a 32 bit latch register, one byte from each plane. This latch register, is not directly accessible from the host CPU; rather it can be used as data for the various write operations. The latch register retains its value until the next read and thus may be used with more than one write operation.
The two read modes, simply called Read Mode 0-1 based on the value of the Read Mode field are:
- Read Mode 0:
Read Mode 0 is used to read one byte from a single plane of display memory. The plane read is the value of the Read Map Selectfield. In order to read a single pixel's value in planar modes, four read operations must be performed, one for each plane. If more than one bytes worth of data is being read from the screen it is recommended that you read it a plane at a time instead of having to perform four I/O operations to the Read Map Select field for each byte, as this will allow the use of faster string copy instructions and reduce the number I/O operations performed. - Read Mode 1:
Read Mode 1 is used to perform comparisons against a reference color, specified by the Color Compare field. If a bit is set in theColor Don't Care field then the corresponding color plane is considered for by the comparison, otherwise it is ignored. Each bit in the returned result represents one comparison between the reference color from the Color Compare field, with the bit being set if the comparison is true. This mode is mainly used by flood fill algorithms that fill an area of a specific color, as it requires 1/4 the number of reads to determine the area that needs to be filled in addition to the additional work done by the comparison. Also an efficient "search and replace" operation that replaces one color with another can be performed when this mode is combined with Write Mode 3.
The VGA has four write modes, selected by the Write Mode field. This controls how the write operation and host data affect the display memory. The VGA, depending on the Write Mode field performs up to five distinct operations before the write affects display memory. Note that not all write modes use all of pipelined stages in the write hardware, and others use some of the pipelined stages in different ways.
The first of these allows the VGA hardware to perform a bitwise rotation on the data written from the host. This is accomplished via a barrel rotator that rotates the bits to the right by the number of positions specified by the Rotate Count field. This performs the same operation as the 8086 ROR instruction, shifting bits to the right (from bit 7 towards bit 0.) with the bit shifted out of position 0 being "rolled" into position 7. Note that if the rotate count field is zero then no rotation is performed.
The second uses the Enable Set/Reset and Set/Reset fields. These fields can provide an additional data source in addition to the data written and the latched value from the last read operation performed. Normally, data from the host is replicated four times, one for each plane. In this stage, a 1 bit in the Enable Set/Reset field will cause the corresponding bit plane to be replaced by the bit value in the corresponding Set/Reset field location, replicated 8 times to fill the byte, giving it either the value 00000000b or 11111111b. If the Enable Set/Reset field for a given plane is 0 then the host data byte is used instead. Note that in some write modes, the host data byte is used for other purposes, and the set/reset register is always used as data, and in other modes the set/reset mechanism is not used at all.
The third stage performs logical operations between the host data, which has been split into four planes and is now 32-bits wide, and the latch register, which provides a second 32-bit operand. The Logical Operation field selects the operation that this stage performs. The four possibilities are: NOP (the host data is passed directly through, performing no operation), AND (the data is logically ANDed with the latched data.), OR (the data is logically ORed with the latched data), and XOR (the data is logically XORed with the latched data.) The result of this operation is then passed on. whilst the latched data remains unchanged, available for use in successive operations.
In the fourth stage, individual bits may be selected from the result or copied from the latch register. Each bit of the Bit Mask field determines whether the corresponding bits in each plane are the result of the previous step or are copied directly from the latch register. This allows the host CPU to modify only a single bit, by first performing a dummy read to fill the latch register
The fifth stage allows specification of what planes, if any a write operation affects, via the Memory Plane Write Enable field. The four bits in this field determine whether or not the write affects the corresponding plane If the a planes bit is 1 then the data from the previous step will be written to display memory, otherwise the display buffer location in that plane will remain unchanged.
The four write modes, of which the current one is set by writing to the Write Mode field The four write modes, simply called write modes 0-3, based on the value of the Write Mode field are:
- Write Mode 0:
Write Mode 0 is the standard and most general write mode. While the other write modes are designed to perform a specific task, this mode can be used to perform most tasks as all five operations are performed on the data. The data byte from the host is first rotated as specified by the Rotate Count field, then is replicated across all four planes. Then the Enable Set/Reset field selects which planes will receive their values from the host data and which will receive their data from that plane's Set/Reset field location. Then the operation specified by the Logical Operation field is performed on the resulting data and the data in the read latches. The Bit Maskfield is then used to select between the resulting data and data from the latch register. Finally, the resulting data is written to the display memory planes enabled in the Memory Plane Write Enable field. - Write Mode 1:
Write Mode 1 is used to transfer the data in the latches register directly to the screen, affected only by the Memory Plane Write Enable field. This can facilitate rapid transfer of data on byte boundaries from one area of video memory to another or filling areas of the display with a pattern of 8 pixels. When Write Mode 0 is used with the Bit Mask field set to 00000000b the operation of the hardware is identical to this mode, although it is entirely possible that this mode is faster on some cards. - Write Mode 2:
Write Mode 2 is used to unpack a pixel value packed into the lower 4 bits of the host data byte into the 4 display planes. In the byte from the host, the bit representing each plane will be replicated across all 8 bits of the corresponding planes. Then the operation specified by the Logical Operation field is performed on the resulting data and the data in the read latches. The Bit Mask field is then used to select between the resulting data and data from the latch register. Finally, the resulting data is written to the display memory planes enabled in the Memory Plane Write Enable field. - Write Mode 3:
Write Mode 3 is used when the color written is fairly constant but the Bit Mask field needs to be changed frequently, such as when drawing single color lines or text. The value of the Set/Reset field is expanded as if the Enable Set/Reset field were set to 1111b, regardless of its actual value. The host data is first rotated as specified by the Rotate Count field, then is ANDed with the Bit Mask field. The resulting value is used where the Bit Mask field normally would be used, selecting data from either the expansion of the Set/Reset field or the latch register. Finally, the resulting data is written to the display memory planes enabled in the Memory Plane Write Enable field.