Address translation

From BeebWiki
Jump to: navigation, search

Address translation is the complex transformation performed by the BBC Micro to map the address spaces of the CPU and CRTC onto the address space of the DRAM array.

Contents

Rationale

Having to be carried out in real time for every bus cycle, the address translation must be done in hardware. Although the task was a potential candidate for a ULA, Acorn chose to implement it in SSI logic. Either way it would have significantly added to the cost of the BBC Micro. Such translation is justified because:

  • the CPU, video ULA and Teletext chip need for memory to be presented in different ways;
  • for speed the display must be scrolled in hardware, which means a virtual address space must be created to 'wrap around' the display memory area;
  • the display memory area, which varies in size, must be moved to the end of RAM. This is because:
    • user memory, memory not allocated to the system or the display, should be a contiguous block, and:
    • changes of MODE should affect BASIC and other languages as little as possible;
    • this also makes it more convenient to determine when to 'wrap around' the display address.

Progression

On the Electron, address translation becomes part and parcel of the ULA's video display activities. This cut-down machine does away with hardware scrolling and Teletext, but the display memory layout is identical to the BBC Micro's MODEs 0 to 6.

The Model B+ expands the subsystem to map 32K of shadow RAM into the address space. The display memory can optionally be moved to shadow RAM, making it visible only to the CRTC and the VDU routines in the MOS, and allowing user memory to extend to the end of main RAM. (For compatibility with games and other programs that update display memory directly, this shadow display can be turned off.)

From the profits of the BBC Micro, Acorn were able to commission several new ULAs for the Master series, one of which implements an advanced memory management service backward-compatible with the Model B and B+.

DRAM address space

The DRAM address bus is a 7 bit bus of lines connecting the address translation buffers to the DRAM memory banks. It is a diplexed bus that runs at 8 MHz, transmitting two addresses for each data transfer at 4 MHz.

The first seven bits transmitted form the row address (ROW); the CRTC visits all rows several times a millisecond and refreshes the DRAM array automatically. The last seven bits form the column address (COL).

These fourteen bits, together with a signal line from IC 28 pin 6 to the north pin of S25, form a 15 bit address space. The bits of the address space are interpreted as follows:

DA14 DA13..DA7 DA6..DA0
Bank select Column address Row address

On the Model A, there is only one DRAM bank as standard so DA14 is unused.

The read/~write line

The read/~write line, supplying the DRAM array, is gated by the buffer ICs parallel to the DRAM address bus. It is only lowered in CPU mode, when the column address is transmitted, and the CPU is writing to RAM (R/~W and A15 both low.) When the CPU is writing to the video ULA registers, the DRAM arrays are disabled but the R/~W line is also held high to prevent spurious writes.

Address translation modes

There are three address translation modes. The BBC micro rapidly alternates between CPU and one of the others, depending on the screen MODE. While reprogramming the CRTC to set the screen MODE, the MOS sets one of its address outputs (MA13) high or low to select the translation mode.

The translation hardware does use two outputs of the addressable latch IC 32, attached to the system VIA, but only in HI RES mode to set the size of the display memory area.

Mode DA14 DA13 DA12 DA11 DA10 DA9 DA8 DA7 DA6 DA5 DA4 DA3 DA2 DA1 DA0
CPU A14 A13 A12 A11 A10 A9 A8 A7 A6 A5 A4 A3 A2 A1 A0
TTX VDU AA3 1 1 1 1 MA9 MA8 MA7 MA6* MA5 MA4 MA3 MA2 MA1 MA0
HI RES AA3 AA2 AA1 AA0 MA7 MA6 MA5 MA4 MA3 MA2 MA1 MA0 RA2 RA1 RA0

CPU

When the 2 MHz monotonic clock is low, the CPU has control of the DRAM data bus. No translation is carried out, thus A0..A14 maps directly to DA0..DA14 of the DRAM address space.

CPU ROW is gated by IC 12 and CPU COL by IC 13.

TTX VDU

This mode is selected when the 2 MHz clock, and the CRTC's MA13 output (TTX VDU) are high. This mode delivers data to the SAA 5050 Teletext generator to display MODE 7. The CRTC's row address (RA) lines are ignored as the SAA 5050 contains its own scanline counter, and expects the same 40 bytes to be presented on each scanline of the text row. Thus:

DA14 DA13..DA10 DA9..DA7 DA6 DA5..DA0
AA3 all 1 MA9..7 MA6 ^ ~1MHz MA5..MA0

In MODE 7 the MOS ensures that MA11 is set and MA12 clear, so that AA3 is 1. This arrangement automatically provides hardware scrolling as MA10, raised on overflow, is ignored and all addresses are in the range &7C00 to &7FFF. By toggling AA3 an application can instantly switch to a second screen at &3C00 to &3FFF.

The MA6 line is XORed with the 1 MHz monotonic clock, so two distinct bytes are fetched in TTX VDU mode every microsecond. The buffer IC 15 delivers one to the SAA 5050 and discards the other. Therefore 80 of the 128 DRAM rows are refreshed per text row, meaning the longest any row goes without a refresh is 640 μs. Without toggling MA6 it would be nearly 2 ms.

TTX ROW is gated by IC 10 and TTX COL by IC 11.

HI RES

This mode is selected when the 2 MHz clock is high and MA13 (TTX VDU) is low. This mode delivers data to the video ULA to display MODEs 0 to 6. The mapping of row and memory address lines to the DRAM address space was designed for efficient printing of text. Normally a character can be displayed by writing 8, 16 or 32 consecutive bytes. Thus:

DA14..DA11 DA10..DA3 DA2..DA0
AA3..AA0 MA7..MA0 RA2..RA0

where AA3..AA0, the adjusted address, is MA11..MA8 with corrections to stay within the display memory area.

(In MODEs 3 and 6, RA3 going high forces DISEN low and so the bottom two scanlines of each text row are blank.)

HI RES ROW is gated by IC 8 and HI RES COL by IC 9.

Calculation of the adjusted address

When MA12 goes high this corresponds to the current display address overrunning the end of RAM at &8000. In order to implement hardware scrolling the address must be 'wrapped around' to restart from the beginning of display memory. In the bitmapped MODEs this is done in hardware in several stages:

  • State: the addressable latch IC 32 is programmed with the encoded display size;
  • Decode: in case of overflow, four logic gates decode the state into a one's-complemented number of 2K units;
  • Subtract: a quad adder IC 39 subtracts this number from the relevant address lines;
  • Diplex: the result is fed to an address translation buffer IC 9 and diplexed onto the address bus as normal.

Outputs 4 and 5 of the addressable latch, labelled C0 and C1, select the size of the display memory. From these the decoder generates outputs, to be fed into the B side of the quad adder, as follows:-

Inputs Outputs Meanings
MA12 C1 C0 B4 B3 B2 B1 Amount to subtract Restart address MODEs
0 x x 1 1 1 1 0 n/a 0..6
1 0 0 0 1 1 1 &4000 &4000 3
1 0 1 1 0 1 1 &2000 &6000 6
1 1 0 0 1 0 1 &5000 &3000 0,1,2
1 1 1 1 0 1 0 &2800 &5800 4,5

-- beardo 16:38, 20 March 2007 (UTC)