C64DTV Programming Guide

From C64-Wiki
Jump to navigationJump to search

About this document[edit | edit source]

This is the C64DTV programming guide. Be bold and add clarifications or additional explanations here.

The original document was written by Jeri and can be found here (PDF). The PDF was converted to a wiki article in the Picobay DTV Hacking Wiki which went offline in 2013. You can still have a look at it via the Wayback Engine.

Terminology/Notation[edit | edit source]

$xxxx
memory address as seen by the CPU
$xxxxxx
physical memory address in RAM/ROM
bank
a memory window. In context with bank switching a memory window where mapping applies. Example: For the DTV CPU, $0000-$3FFF is RAM bank 0.
segment
a memory area being mapped into a bank. Example: In default configuration, RAM segment 0 ($000000-$003FFF) is mapped to the DTV CPU's RAM bank 0 ($0000-$3FFF).

DTV2/3 Basic Block Diagram[edit | edit source]

+-------+--+---+        +---------+
|65DTV02|->|MMU|---+--->|   VIC   | 
+-------+--+---+   |    +---------+
                   |    | Sprites |
+-----+            |    +---------+
| SID |<-----------+
+-----+            |    +---------+
                   +--->|   DMA   |
+--------+-----+   |    +---------+
|Keyboard| CIA |<--+    | Blitter |
+--------+-----+   |    +---------+
                   V         
        +-----------------------+
        |SDRAM/Memory Controller|--->
        +-----------------------+

Memory layout[edit | edit source]

6502 map at reset[edit | edit source]

This is memory as seen by the CPU after reset. This layout may be changed using the C64 PLA ($1), the DTV's memory mapper, or the DTV's register file/segment mapper.

$FFFF   +---------------+
        |               |
$F800   |     KERNEL    |
        |       &       |
        |     EDITOR    |
        |   FROM FLASH  |
        |$00E000-$00FFFF|
$E000   +---------------+
        |  COLOR NYBS   |
        |  I/O & CHARS  |
$D000   +---------------+
             .  .  .
$BFFF   +---------------+ 
        |               | 
        |     BASIC     | 
        |               | 
        |   FROM FLASH  |
        |$00A000-$00BFFF| 
$A000   +---------------+ 
             .  .  .
$01FF   +---------------+
        |               |
        |     Stack     |
        |               |
$0100   +---------------+
             .  .  . 
$0001   +---------------+
        |    6510 IO    |
$0000   +---------------+

IO REGISTERS[edit | edit source]

        +---------------+
$DF00   |     I/O-2     | EXTERNAL I/O 
        +---------------+
$DE00   |     I/O-1     | EXTERNAL I/O 
        +---------------+
$DD00   |     CIA-2     | SERIAL/USR PRT
        +---------------+
$DC00   |     CIA-1     | KEYBRD/JOYSTCK
        +---------------+
        |    COLOR      | COLOR MATRIX 
$D800   |   FROM RAM    |
        |$01D800-$01DBFF|
        +---------------+
$D400   |      SID      | AUDIO
        +---------------+
$D300   |  DMA/BLITTER* | DMA CONTROLLER/BLITTER
        +---------------+
$D200   |    PALETTE*   | LUMA/CHROMA 
        +---------------+ 
$D100   |      MMU*     | MEMORY MAPPER
        +---------------+ 
$D000   |      VIC      | VIDEO 
        +---------------+ 
*Extended control must be enabled to address these registers

ROM LAYOUT[edit | edit source]

This is the (Flash)ROM layout of a DTV.

            .   .   .
$1D0000 +---------------+
        |               |
        |     FILES     |
        |               |
$014000 +---------------+
        |   DIRECTORY   |
$010000 +---------------+
        |               |
        |    KERNAL     |
        |               |
$00E000 +---------------+
        |      CPU      |
        |   CHARACTER   |
        |      SET      |
$00D000 +---------------+
            .   .   .
$00BFFF +---------------+
        |               |
        |     BASIC     |
        |               |
$00A000 +---------------+
        |      VIC      |
        |   CHARACTER   |
        |     SET 2     |
$009000 +---------------+
            .   .   .
$001FFF +---------------+
        |      VIC      |
        |   CHARACTER   |
        |     SET 1     |
$001000 +---------------+

At $010000 the directory of files stored in upper Flash (accessible using device number 1) is stored in default setup. This is not determined by hardware but by software/kernal - see DTV Flash File System for details.

RAM LAYOUT[edit | edit source]

This is pretty the same as with the original C64. However, the original VICII has its own colormem additional to the C64's 64k - this memory is mapped at $01D800 in the DTV's RAM.

$01DBFF +---------------+
        |               |
        | COLOR MATRIX  |
        |               |
$01D800 +---------------+

EXTENDED SID REGISTERS[edit | edit source]

$D41E Writes to voice 1s upper 8 bits of waveform accumulator.
$D41F Writes to voice 2s 8bit envelope generator.

Writing to voice 1s waveform accumulator when frequency is set to 0 can be used for 8 bit digital sample playback. The accumulator may also be set to a non-zero frequency and written to for compressed sample playback.

Saw tooth:

The MSB will set the direction of the ramp.
0 = ascending
1 = descending

The accumulator frequency sets the angle off ramp.

   2   ___4
  /\  /   |
 /  \/ 3  | 
/         |____  
1

Complex waveforms can be generated with fewer data points

  1. Accumulator is loaded with %00000000 (ascending MSB = 0) and frequency is set to fast rise time.
  2. Accumulator is loaded with %11111111 (descending MSB = 1) and slower frequency.
  3. Accumulator is loaded with %11111111 and frequency is set to 0.
  4. Accumulator is loaded with %00000000 and frequency is set to 0.
  • Noise: Toggling bit[3] from 0 to 1 will advance the noise LFSR register.
  • During Channel 2 Sync modulation: Toggling accumulator bit[7] from 1 to 0 will sync modulate voice2.
  • During Channel 2 Ring modulation: Setting bit[7] = 1 in accumulator will make channel 2 waveform descend.

Writing to voice 2s envelope generator.

       2
  1 /\/\    3
 /\/    \___/\__
/               \_
  • During attack:
    • Writing a lower value than current attack level will start attack over at the lower position (see point 1)
    • Writing %11111111 will start decay state.
  • During Decay:
    • Writing a higher value than current decay level will start decay over at the higher position (see point 2)
    • Writing sustain value will change to sustain state.
    • Writing a value less than sustain level causes envelope to decay towards zero at decay rate.
  • During Sustain:
    • Writing a higher value than sustain will start decay again. (see point 3).
    • Writing a lower value than decay will cause envelope to decay towards zero at decay rate.
  • During Release:
    • Writing higher value than current release will start release at a higher value.

VIC FUNCTION BLOCKS[edit | edit source]

The VIC has 4 major functional blocks: Address generator, pixel shifters, color decoder and color/character data line buffer.

The address generator forms all of address to be used fetches of graphics, character pointers and color data.

Pixel shifters take fetched data and shifts out 1,2,4 or 8 bits per dot clock (or pixel on screen)

The color decoder maps colors to pixel data that is shifted out of the pixel shifters. Color data may come from fetched color matrix, character matrix or color registers.

Color/character line buffer is a 40 x 16 bit memory that stores the character and color data every bad line. This memory is read back over the next 8 lines and used with the color encoder.

EXTENDED VIC REGISTERS[edit | edit source]

$D036 53302  Color Bank Low
             (See address generator) 
$D037 53303  Color Bank High
             (See address generator)
$D038 53304  Linear Count A modulo low
             (See address generator)
$D039 53305  
      Bits[3:0]      
             Linear Count A modulo 
             high
             (See address generator)
$D03A 53306  Linear Count A Start low
             (See address generator)
$D03B 53307  Linear Count A Start 
             middle
             (See address generator)
$D03C 53308       
      Bit[0] Linear addressing when set
      Bit[1] Border off when set
      Bit[2] High color when set 
             (Extended color decoder 
             mode)
      Bit[3] Overscan For linear modes
             50 columns NTSC
             47 columns PAL
             This mode is severely 
             broken.  It ignores 
             modulos unless disabled 
             around cycle 57 and messes 
             with sprite fetches. 
      Bit[4] ColorRAM Fetch Disable
             Repeats value that are in
             line buffer.  Line buffer
             is cleared during VBlank.
      Bit[5] CPU bad line Disable 
             (bad lines are emulated 
             for CPU for compatibility)
      Bit[6] Chunky Enable
             (See color decoder and
             address generator)

$D03D 53309  Graphics fetch bank
             (used for all classic VIC
             graphics fetches) 
$D03F 53311
      Bit[0] Enable extended feature 
             registers when set.  This
             will also enable 256 color
             $D020, $D021, $D022, 
             $D023, $D024 
             Registers set under this 
             Mode will remain after bit
             is cleared.     
      Bit[1] Setting disables extended
             modes until next reset.
$D040 53312  
      Bit[0] PAL line timing when set
             (63 cycles in PAL 65 in 
             NTSC)
      Bit[1] Burst phase alternate when
             set
        
      Bit[2] V1 DTV Palette 
             compatibility when set

$D041 53313  Burst rate modulus high
             Default = 28
$D042 53314  Burst rate modulus middle
             Default = 19
$D043 53315  Burst rate modulus low 
             Default = 120
        
Fout = SysClk * N/16777216

Where N is the modulo value and Fout is desired burst frequency

NTSC/32.64mhz SysClk
3.579545mhz / 0.00000194549560546875=
1839914.2048627450980392156862745
NTSC/32.64mhz = 1C132A

PAL/31.36mhz SysCLk
4.433619mhz / 0.00000186920166015625=
2371931.8757877551020408163265306
PAL/31.36mhz Modulus = 24315B  

Ntsc/32.7272mhz SysClk
0.0000019506931304931640625
1835011.8447872106382458627685839
NTSC/32.7272mhz Modulus = $1C0000

PAL 31.5279mhz SysClk
0.0000018792092800140380859375
2359300.2903683404222926360461686
PAL/31.5279mhz  Modulus = 240000 


$D044 53316  
      Bits[6:0]
             During reads
             CPU Cycle

             Writes IRQ trigger cycle
            NTSC = 0->64
             PAL = 0->55 & 58->64
             (PAL Skips 2 cycles)
             Default = 64
$D045 53317  
      Bits[5:0]
             Linear Count A Start high
             (See address generator)
$D046 53318  Linear Count A Step
             (See address generator)
$D047 53319  Linear Count B modulo low
             (See address generator)
$D048 53320       
      Bits[3:0]
             Linear Count B modulo high
             (See address generator)
$D049 53321  Linear Count B start low
             (See address generator)
$D04A 53322  Linear Count B start Mid
             (See address generator)
$D04B 53323  
      Bits[5:0]
             Linear Count B start high
             (See address generator)
$D04C 53324  Linear Plane B Step
             (See address generator)
$D04D 53325  
      Bits[5:0]
             Sprite Bank 

$D04E 53326  Scan line timing adjust.
             Adds about 25ns per count

             NTSC/32.64mhz = $D
             PAL/31.36mhz  = $5

             NTSC 32.72mhz = $0
             PAL  32.5279mhz = $0
$D04F 53327
      Bit[1:0]
             Saturation 
               00 lowest 
               11 highest
      Bit[2] Burst lock to line 
             PAL/NTSC
      Bit[3] Burst lock to line with  
             negative phase walk     
        

VIC ADDRESS GENERATION[edit | edit source]

Sprite Addresses
Sprite Pointer Fetch:
(Sprite Bank & Character Matrix)
Sprite Graphics Fetch:
(Sprite Bank & Pointer & DMA Count)


Linear addressing = 0
Graphics Fetches:
(GfxBank[5:0] & VIC Address[15:0])
Character Fetches:
(LinearCountA[21:16] & VIC Address[15:0])
Color Fetches:
(ColorBank[11:0] & Matrix[9:0])

VIC Addresses are standard 64k addressing modes set with MCM/BMM/ECM

Set linear count to step0 and set Start Address A[21:16] to desired character bank. Linear A counter will count 40 steps and add 1 modulus per active scan line.


Linear addressing = 1
Color Fetch Disable = 0
Chunky Enable = 0
Graphics Fetches:
(Plane A Linear Address[21:0])
Character Fetches:
(Plane B Linear Address[21:0])
Color Fetches:
(ColorBank[11:0] & Matrix[9:0])

COLOR DECODER[edit | edit source]

ECM = 0 BMM = 0 MCM = 0 HIGHCOLOR = 1
Plane A = 0 (8bit background color 0)
Plane A = 1 (8bit color data)
ECM = 1 BMM = 0 MCM = 0 HIGHCOLOR = 1
Plane A = 0
      Character data [7:6]
      +-----------------------------+
      |00 = 8bit background color 0 |
      |01 = 8bit background color 1 |
      |10 = 8bit background color 2 |
      |11 = 8bit background color 3 |
      +-----------------------------+
Plane A = 1 (8bit color data)
ECM = 0 BMM = 0 MCM = 1 HIGHCOLOR = 1
Color data[3] = 0
Plane A = 0 (8bit background 0)
Plane A = 1 Color[7:4]0Color[2:0]
Color data[3] = 1
      Plane A pixels only  
      +-----------------------------+
      |00 = 8bit background color 0 |
      |01 = 8bit background color 1 |
      |10 = 8bit background color 2 |
      |11 = Color[7:4] 0 Color[2:0] |
      +-----------------------------+
ECM = 0 BMM = 1 MCM = 1 HIGHCOLOR = 1
Plane A = 00 (8bit background 0)
Plane A = 01 (0000 Character[7:4])
Plane A = 10 (0000 Character[3:0])
Plane A = 11 (8bit color data)

Sixs FRED MODE - 8bpp Packed Bitmap[edit | edit source]

ECM = 1 BMM = 1 MCM = 1 HIGHCOLOR = 1
8 bit pixel is made up of
(ColorRam[3:0],PlaneBShifter[1:0], PlaneAShifter[1:0])

One could think of this mode as FLI with no cpu overhead and a re-definable palette, but it is actually much more powerful. It is a cellular mode, with 4x8 cells. Each pixel is 2 hires pixels wide. Each 4x8 cell can contain any of 16 colors, the downside being that those 16 colors have to have the same high nibble, which is determined by the ColorRam for that cell. Thus, if ColorRam is $00, you can use $00-$0f in that cell, $01 you can use $10-$1f in that cell. The lower nibble is set as shown above, bits 0-1 being from Plane A, bits 2-3 from Plane B. So if Color Ram is $04, the byte in Plane A is %10101100, and the byte in Plane B is %00011010, the pixel colors will be $48,$49,$4e,$42.

See DTV FRED Mode for details.

Sixs FRED MODE2[edit | edit source]

ECM = '1' BMM = '1' MCM = '1'
HIGHCOLOR = 0 LinearAddressing = '1'
8 bit pixel is made up of
(PlaneBShifter[1:0],ColorRam[3:2],PlaneAShifter[1:0],ColorRam[1:0])

See DTV FRED Mode 2 for details.

Two Plane Bitmap[edit | edit source]

ECM = 1 BMM = 1 MCM = 0 HIGHCOLOR = 1
LinearAddress = 1
Plane A = 0 Plane B = 0 (Background 0)
Plane A = 0 Plane B = 1 (0000 Color[7:4])
Plane A = 1 Plane B = 0 (0000 Color[3:0])
Plane A = 1 Plane B = 1 (Background 1)

CHUNKY 8BPP Bitmap[edit | edit source]

ECM = 1 BMM = 0 MCM = 1 HIGHCOLOR = 1
ColorFetchDisable = 1 LinearAddress = 1
ChunkyEnable = 1

Chunky mode displays 8 8bit pixels per CPU cycle. The pixels come from counter B. To set up a linear video frame buffer the step size must be set to 8.

8BPP Pixel Cell[edit | edit source]

ECM = 1 BMM = 0 MCM = 1 HIGHCOLOR = 1
ColorFetchDisable = 0 LinearAddress = 1
ChunkyEnable = 1

Pixel cell mode can be thought of as a text mode that uses a 8x8 pixel 8BPP font. The characters come from counter A and the font (or "cells") from counter B. Counter B step and modulo should be set to 0, counter A modulo to 0 and counter A step to 1 for normal operation. ColorRAM is unused. The pixel data in the cells is organized as:

  Cell 1 Data      Cell 2 Data
0 1 2 3 4 5 6 7    64   ...     72 
8             16
      .... 
              
55            63

VIC ADDRESS GENERATOR[edit | edit source]

The VIC has three cycles to fetch data per 8 pixels displayed

Cycle 1: Character Fetch/Counter A
Cycle 2: Color Fetch/DMA/Blitter
Cycle 3: Graphic Fetch / Counter B
Cycle 4: CPU Access

Addresses for each of the cycles are generated with counters (some with modulus).

Counter A: 22 bits with start, modulo and step
Counter B: 22 bits with start, modulo and step
RowCounter: 3 bits that count the lines from the last bad line. It terminates at 7
Matrixcount: 10 bits that increments every character read during bad lines

Character Fetch Addresses[edit | edit source]

Linear count = 0 ChunkyEn = Dont Care ColorDisable = Dont care

Used during normal legacy vic operation to read character matrix. Linear count A can be enabled to change banks during screen fetches or set to a constant for a bank.

Address = LinearCountA(21 downto 16) & pa & vm & matrix_counter


Linear count = 1 ChunkyEn = 1 ColorDisable = 0

Used with 8bpp pixel cell mode.

Address = LinearCountA


Linear count = 1 ChunkyEn = Dont Care ColorDisable = 1

Used with plane type bitmaps and chunky 8bpp

Address = LinearCountB

Color Fetch Addresses[edit | edit source]

ChunkyEn = 1 ColorDisable = 0

I cant remember why this is here at the moment

Address = LinearCountA


ChunkyEn = 0

This addressing mode is used during legacy VIC addressing.

Address = ColorBankHigh & ColorBankLow & matrix_counter

Graphics Fetch[edit | edit source]

bmm = 0 ecm = 0 LinearAddressing = 0
Address = GraphicsBank & pa & cb & CharacterPointer & row_counter
bmm = 0 ecm = 1 LinearAddressing = 0
Address = GraphicsBank & pa & cb & "00" & CharacterPointer(5 downto 0) & row_counter
bmm = 1 LinearAddressing = 0
Address = GraphicsBank & pa & cb(2) & matrix_counter & row_counter


ChunkyEn = 1 ColorDisable = 0

This mode is used for 8bpp pixel cell mode.

Address = LinearCountB(21 downto 14) & CharacterPointer & row_counter (& pixel_counter)


ChunkyEn = 1 ColorDisable = 1

Used for 8bpp bitmap mode. Note the inverted linearCountB. This keeps character fetch and this fetch 4 bytes apart with the same counter.

Address = LinearCountB(21 downto 3) & not LinearCountB(2) & LinearCountB(1 downto 0)

SETTING VIDEO STANDARDS[edit | edit source]

The DTV allows individual control of different components of PAL and NTSC. The components can be mixed and matched to create NTSC, NTSC(J) and PAL. Control registers are:

$D040 53312  
      Bit[0] PAL line timing when set
      Bit[1] Burst alternate when set
      (Other bits are in this
       this register) 
$D041 53313 Burst rate modulus high
$D042 53314 Burst rate modulus middle
$D043 53315 Burst rate modulus low
$D04E 53326 Scan line timing adjust
$D04F 53327 Scan line phase 
            relationship      

$D040 53312  
      Bit[0] PAL line timing when set
      Bit[1] Burst alternate when set

Bit[0] adjusts PAL line timing to have 63 CPU cycles horizontal proper scan rate with a 31.xxx mhz crystal. When cleared there will be NTSC scan line timing to have 65 cycles and a proper scan rate with a 32.xxxmhz crystal.

Bit [1] enables PAL backwards 1/4 phase backwards burst walk per scan line and 180deg alternation. NTSC mode locks 180 drift per scan line.

$D041 53313 Burst rate modulus high
$D042 53314 Burst rate modulus middle
$D043 53315 Burst rate modulus low

Color is generated with reference to the burst frequency. The burst modulus registers set a fractional digital synthesizer.

Fout = SysClk * N/16777216

Where N is the modulo value and Fout is desired burst frequency

NTSC/32.64mhz
SysClk 32.64/16777216 = 0.0000019454956054687
Burst 3.579545mhz / 0.00000194549560546875 = 1839914.2048627450980392156862745
NTSC Modulus = $1C132A
PAL/31.36mhz
SysCLk
Burst 4.433619mhz / 0.00000186920166015625 = 2371931.8757877551020408163265306
PAL Modulus $24315B
Ntsc 32.7272
0.0000019506931304931640625
1835011.8447872106382458627685839
NTSC Modulus = $1C0000
PAL 31.5279mhz xtal
0.0000018792092800140380859375
2359300.2903683404222926360461686
PAL Modulus = 240000
$D04E 53326 Scan line timing adjust.

Color information in PAL and NTSC have a precise relationship with horizontal timing. The lower nibble of this register will add ~20ns(crystal dependant) per value to the scan line. Adjust this to have stable color lock

NTSC/32.64mhz = $D
PAL/31.36mhz = $5
NTSC 32.72mhz = $0
PAL 32.5279mhz = $0
$D04F 53327 Scan line phase relationship

The PAL video standard alternates the color information 180 degrees every other scan line and NTSC maintains a constant phase relationship. Phase alternating relationship can be adjusted in 22.5 deg steps relative to burst and relative every other line with $D04F. Use this to fine tune hue and interline color.

DIGITAL VIDEO OUTPUT[edit | edit source]

For a possible hardware mod adding digital output, you can use the 4 luma bits without a problem, but the 4 chroma bits need special initialization. The color subcarrier for chroma is generated by a digital frequency synthesizer(DFS) and phase shifted by adding the 4 bit value in color look up table. To get meaningful values from chroma you'll need to precisely stop the DFS when outputing "0000" and set saturation to maximum ("11").

Stopping DFS at "000":

  • Set all three burst rate registers to "00000000".
  • Set PhaseAlternate to '0' .
  • Set Burst Lock to '1' (Bit[2] $d04f) until after burst on odd line(LSB Raster = '0')
  • Set burst lock to '0' before burst on even line.

The DFS is now flooded with 0's, stopped and you'll now see chroma values in the CLUT reflected in the chroma pins. H and V Sync can be extracted with any low cost sync separator (LM1881). Now you can hook it to a digital RGB monitor, R2R ladder or RAMDAC.

OVERSCAN REGISTER[edit | edit source]

An attempt to add overscan happened a day or two before final tapeout, but there wasn't enough time to complete it. The register was not removed just in case something useful could be done with it (glitches or not)..

What is known about the register:

  • Overscan starts fetching data from cycle 8 -> 57 in linear mode.
  • Enabling overscan for a minimum of 1 cycle after cycle57 and before cycle13 will force DisplayWindowEnable on. DisplayWindowEnable = on in blanking will cause character, color and graphics addresses instead of sprites. It will also start the bitmap pixel shifter.
  • Matrix base is cleared, linear count A&B are loaded on line 10 instead of 48.
  • Will set EnableDisplay anywhere in display causing MatrixCounter to increment between cycle 14 and 53. If in chunky LinearCountA will also increment by step rate.
  • RowCounter will increment on cycle 58
  • Sprites can collide in border area.

DMA and Blitter[edit | edit source]

DMA REGISTERS[edit | edit source]

$D300 Source [7:0]   (Low)
$D301 Source [15:8]  (Middle)
$D302 Source [23:16] (High)
      Bits[23:22] 00 = ROM
                  01 = RAM
                  10 = RAM + Registers
$D303 Destination[7:0]  (Low)
$D304 Destination[15:8] (Middle)
$D305 Destination[23:16] (High)
      Bits[23:22] 00 = ROM
                  01 = RAM
                  10 = RAM + Registers
$D306 Source Step[7:0]
$D307 Source Step[15:8]
$D308 Destination Step[7:0]
$D309 Destination Step[15:8]
$D30A DMA Length[7:0]
$D30B DMA Length[15:8]
$D30C Source Modulo[7:0]
$D30D Source Modulo[15:8]
$D30E Destination Modulo[7:0]
$D30F Destination Modulo[15:8]
$D310 Source Line Length[7:0]
$D311 Source Line Length[15:8]
$D312 Destination Line Length[7:0]
$D313 Destination Line Length[15:0]
$D31D ClearIRQ[0] Write 1 to clear IRQ
      SourceContinue[1]
      DestinationContinue[3]  Continues
                 Last address when set
$D31E Source Modulo Enable[0] when set
      Destination Modulo Enable[1]
$D31F Bit[0]   Force Start DMA when set
      Bit[1]   Swaps source mem with 
               destination mem when set.
               Copies source mem to
               destination mem when unset
      Bit[2]   Source Direction 
               Positive when set
      Bit[3]   Destination Direction 
               Positive when set
      Bit[4]   VIC IRQ Start enables 
               DMA on VIC IRQ when set
      Bit[5]   Start on blitter done 
               when set
      Bit[6]   VBlank Start when set
      Bit[7]   IRQ Enable Enables DMA 
               Done IRQs when set
   
      During reads
      Bit[0]  DMA Busy
      Bit[1]  IRQ

BLITTER REGISTERS[edit | edit source]

$D320  Source A [7:0]   (Low)
$D321  Source A [15:8]  (Middle)
$D322  Bits[5:0]
       Source A [21:16] (High)
$D323  Source A Modulo[7:0]
$D324  Source A Modulo[15:8]
$D325  Source A Line Length[7:0]
$D326  Source A Line Length[15:8]
$D327  Source A Step (Fractional)
       [7:4] integral part
       [3:0] fractional part
$D328  Source B [7:0]   (Low)
$D329  Source B [15:8]  (Middle)
$D32A  Bits[5:0]
       Source B [21:16] (High)
$D32B  Source B Modulo[7:0]
$D32C  Source B Modulo[15:8]
$D32D  Source B Line Length[7:0]
$D32E  Source B Line Length[15:8]
$D32F  Source B Step (Fractional)
       [7:4] integral part
       [3:0] fractional part
$D330  Destination [7:0]   (Low)
$D331  Destination [15:8]  (Middle)
$D332  Bits[5:0]
       Destination [21:16] (High)
$D333  Destination Modulo[7:0]
$D334  Destination Modulo[15:8]
$D335  Destination Line Length[7:0]
$D336  Destination Line Length[15:8]
$D337  Destination Step (Fractional)
       [7:4] integral part
       [3:0] fractional part
$D338  Blit Length[7:0]  (Low)
$D339  Blit Length[15:0] (high)
 
$D33A  Bit[0] Force Start Strobe when set
       Bit[1] Source A Direction Positive when set
       Bit[2] Source B Direction Positive when set
       Bit[3] Destination Direction Positive when set
       Bit[4] VIC IRQ Start when set
       Bit[5] CIA IRQ Start when set($DCXX CIA)
       Bit[6] V Blank Start when set
       Bit[7] Blitter IRQ Enable when set

$D33B
     Bit[0]   Disable Channel B
              (data into b port of ALU
              is forced to %00000000.
              ALU functions as normal)
     Bit[1]   Write Transparent Data 
              when set
              (Data will be written if
              source a data *IS* 
               %00000000.  This can be
              used with channel b and 
              ALU set to OR to write 
              Data masked by source A.)
              Cycles will be saved if  
              No writes.
     Bit[2]   Write Non Transparent 
              when set
              (Data will be written
              if SourceA fetched data 
              is *NOT* %00000000.  This
              may be used combined with 
              channel b data and/or
              ALU)  Cycles will be 
              Saved if no write.
     Bit[2]==Bit[1]==0: write in any case
 
$D33E  Bit[2:0] Source A right Shift
              000 SourceA Data
              001 LastA[0],SourceA[7:1]
                         ...
              111 LastA[6:0],SourceA[7]
     Bit[5:3] Minterms/ALU
              000 AND
              001 NAND
              010 NOR
              011 OR
              100 XOR
              101 XNOR
              110 ADD A + B
              111 SUB A - B

$D33F  Bit[0] Clear Blitter IRQ
     Bit[1] Source A Continue
     Bit[2] Source B Continue
     Bit[3] Destination Continue
            Restart counters from 
            location they stop during 
            last blit.

     During Reads
     Bit[0] Busy when set
     Bit[1] IRQ when set

BLITTER DATAPATH[edit | edit source]

             Source A       Source B
                 |              |
                 V              V
           +-----------+  +-----------+
           | Comparator|  |DisableGATE|
           | == or !=  |  | %0000000  |
           +-----------+  +-----------+
                 |              |
         +-------+              |
         |       |              |
         |       |              | 
   +-----------+ |              | 
   |Last A Data| |              | 
   +-----------+ |              |
         |       |              |
         |       |              |
         |       V              |
         |  +-------+           |
         +->|Shifter|           |
            +-------+           |
                 |              |
                 V              V
               \-----\______/-----/
                \                /
                 \     ALU      /
                  --------------
                         |
                         V
                    Destination 
                       Data

Last SourceA register stores data from previous access and can be shifted into MSBs of current SourceA. Last SourceA register is cleared at start of DMAs and at the end of each DMA line (when modulus value is applied to SourceA address). This is useful for scrolling video data.

Blitter and DMA length is the total number of bytes to be transferred in one triggered event.

Line length is the length of one contiguous stream of data, before a modulus value is added to address.

Modulus values are added to addresses when the line length for the channel has been reached. This is useful for moving rectangular blocks of data.

Continue bits when set will keep last address value for channel after DMA stops. These need to be set after the first DMA access or addresses will not be set with start value.

IRQ start bits when set will automatically start a DMA access when the IRQ condition is true. These are edge triggered and will only start if in idle state.

Blitter accesses take advantage of the burst access of SDRAM and can only operate on SDRAM. The DMA channel can operate on any RAM, ROM or registers, but does not support burst access (transfers are slower)

The blitter will cache 4 bytes of data and will not access RAM for that channel if new data is not needed. This saves memory bandwidth and allows for constant values.

Disabling Color accesses will give maximum cycles for DMA. Sprite DMA will have still higher priority than blitter.

Bandwidth Examples:
  During 65 cycle line, no sprites, read/write steps of 1,no color fetch and 1 channel reads.

Cycle 0     4 reads
Cycle 1-4   4 writes
Cycle 5     4 reads
Cycle 6-9   4 writes
Cycle 10    4 reads
Cycle 11-14 4 writes
Cycle 15    4 reads
Cycle 16-19 4 writes
Cycle 20    4 reads
Cycle 21-24 4 writes
Cycle 25    4 reads
Cycle 26-29 4 writes
Cycle 30    4 reads
Cycle 31-34 4 writes
Cycle 35    4 reads
Cycle 36-39 4 writes
Cycle 40    4 reads
Cycle 41-44 4 writes
Cycle 45    4 reads
Cycle 46-49 4 writes
Cycle 50    4 reads
Cycle 51-54 4 writes
Cycle 55    4 reads
Cycle 56-59 4 writes
Cycle 60    4 reads
Cycle 61-64 4 writes

Total bytes transferred = 52
Bytes transferable in 262 lines = 13624
(Not counting pre-buffered reads and transparent pixel optimizing)

(~10) 40 X 32 pixel 8bpp BOBs (1280 bytes) can be placed per frame.
(~18) 40 x 32 Pixel Fred1/2 BOBs(1280 bytes.


  During 65-cycle line, no sprite, read/write steps of 1, no color fetch and 2 channels reads.

Cycle 0       4 reads
Cycle 1       4 reads
Cycle 2-5     4 writes
Cycle 6       4 reads
Cycle 7       4 reads
Cycle 8-11    4 writes
Cycle 12      4 reads
Cycle 13      4 reads
Cycle 14-17   4 writes
Cycle 18      4 reads
Cycle 19      4 reads
Cycle 20-24   4 writes
Cycle 25      4 reads
Cycle 26      4 reads
Cycle 27-30   4 writes
Cycle 31      4 reads
Cycle 32      4 reads
Cycle 33-36   4 writes
Cycle 37      4 reads
Cycle 38      4 reads
Cycle 39-42   4 writes
Cycle 43      4 reads
Cycle 44      4 reads
Cycle 45-46   4 writes
Cycle 47      4 reads
Cycle 48      4 reads
Cycle 49-52   4 writes
Cycle 53      4 reads
Cycle 54      4 reads
Cycle 55-58   4 writes
Cycle 59      4 reads
Cycle 60      4 reads
Cycle 61-64   4 writes

Total bytes transferred = 44
Bytes transferable in 262 lines = 11528
(Not counting pre-buffered reads and transparent pixel optimizing)

(~9) 40 X 32 pixel 8bpp BOBs (1280 bytes) can be replaced per frame.
(~16) 40 x 32 Pixel Fred1/2 BOBs(1280 bytes.


Fractional incrementing can be used for scaling data.

Bits[7:4] are whole number of steps.  
Bits[3:0] are fractions of steps per  
          Access

Examples:
Default  %00010000 (step of 1 to 1)
         %00001000 (step of 1 to .5)
         %00000100 (step of 1 to .25)
         %00001100 (step of 1 to .75)
         %00000000 (no step)

Steps less than 1 on source channels can save read accesses. For example source steps of .25 will take 1 read access instead of 4 for the same amount of writes.

It may be advantageous to have a source channel with a slower step than the other source channel when using minterms to transform higher frequency data with low frequency data.

Steps of 0 on read channels will cause the blitter to read(once) the start address and use that value as a constant during the blit operation. Any update to this memory location after blit has started will not be recognized, since the value is in the blitters cache.

Step of 0 on destination channel will cause all writes to the start address. (may not be very useful)

Non-transparency blits write whole bytes to memory if channel A is non zero value (good for placing images on chunky bitmaps). Zero values will save one write cycle.(Very good thing!)

Transparency blits write whole bytes to memory if values in channel a are zero. Combined with channel B and the alu set to OR the reverse of non-transparency can be done affectively only replacing parts of bitmap that had been written before. Non-zero values save one write cycle. (Very good thing!)

On devices using the DTV2 ASIC writes always occur when srcA is fetched regardless of transparency setting - this is a bug.

Reads on blits are 4x faster than writes.

Memory handling[edit | edit source]

MEMORY MAPPER[edit | edit source]

The memory mapper can select which segment the ROMs are fetched from. The ROMs may be fetched from RAM, but still will remain write protected.

  • $D100 Kernal segment
  • $D101 Basic segment
Bits [5:0] Segment number (64k size), defaults to 0
Bits [7:6] 00 = ROM
           01 = RAM

Example: If the CPU accesses the kernal area ($00E000-$00FFFF), physical memory at addr + ((io(0xd100) & 0x3f) << 16) is accessed.

Registers are write only and may only be accessed when extended mode (see $D03F) is active.

Moving Kernal and Basic into SDRAM will allow a 4x speed increase in CPU burst mode.

The registers are aliased every $10 bytes.
Note that the range $D100-$D1FF is also mapped through in to RAM at $D100-$D1FF when accessing the memory mapper.
Reads will read RAM. All writes will write RAM, even if they also write the memory mapper registers.
This is a hardware bug.

CPU EXTENSIONS[edit | edit source]

REGISTER FILE[edit | edit source]

The original 6502 only contained 3 registers (A, X and Y). The DTV now contains 16 registers which can be mapped into A, X and Y. Registers 10 - 15 are dual purpose banking registers and can also be used with ALU operations.

   +-------------+
   |    REG 0    |  DEFAULT ACCUMULATOR 
   +-------------+
   |    REG 1    |  DEFAULT Y REGISTER 
   +-------------+
   |    REG 2    |  DEFAULT X REGISTER 
   +-------------+
   |    REG 3    |  
   |     TO      |  Reserved (not implemented)
   |    Reg 7    |
   +-------------+
   |    REG 8    |  Bank0-3 Access Mode
   +-------------+
   |    REG 9    |  CPU Control
   +-------------+
   |    REG 10   |  BASE PAGE SEGMENT (256 byte segment size)
   +-------------+
   |    REG 11   |  STACK SEGMENT (256 byte segment size)
   +-------------+
   |    REG 12   |  BANK 0 SEGMENT (16k segment size)
   | $0000-$3FFF | 
   +-------------+
   |    REG 13   |  BANK 1 SEGMENT (16k segment size)
   | $4000-$7FFF |
   +-------------+
   |    REG 14   |  BANK 2 SEGMENT (16k segment size)
   | $8000-$BFFF |
   +-------------+
   |    REG 15   |  BANK 3 SEGMENT (16k segment size)
   | $C000-$FFFF |
   +-------------+

CALCULATING PHYSICAL RAM/ROM ADDRESSES[edit | edit source]

If the CPU accesses memory, the physical address gets calculated as follows:

  1. If opcode is zeropage opcode: addr = (addr & 0xff) + (reg10 << 8)
  2. If opcode is stack opcode: addr = (addr & 0xff) + (reg11 << 8)
  3. bank = addr >> 14
  4. addr = (addr & 0x3fff) + (reg[12+bank] << 14)
  5. reg8AccessMode(bank) == ROM? Yes: addr is ROM/FLASH address, done
  6. addr > 0xffff? Yes: Access RAM - No: pass addr to standard MMU/PLA

Some interesting facts follow:

  • STA $01 can be different from STA $0001 (as the latter is no zeropage opcode, thus step 1 does not apply)
  • Zeropage opcodes may access any DTV memory address including I/O registers
  • Addresses may get translated three times before getting the physical address: 1. reg10/11, 2. reg12-15, 3. C64 MMU
  • 6502 I/O ($0/$1) is actually $000000/$000001.

Please note that these translations only apply for the CPU. VIC/DMA/Blitter access physical memory directly.

ACCUMULATOR CONTROL[edit | edit source]

Besides selecting between 16 registers the accumulator may also have separate source and destination register file. This allows the source register to remain constant while only updating the destination. The accumulator may also be pointed at the same register that X or Y is pointing at.

The two byte opcode $32 (SAC) sets the source and destination register file. The immediate value bits [7:4] set the destination and bits [3:0] set the source.

ACCUMULATOR WITH SAME SOURCE AND DEST[edit | edit source]

 +--------+          
 |        |               
 |        V               
 | +-------------+        
 | |    REG 0    |   FROM MEMORY
 | +-------------+	   |
 |        |               |
 |        |               | 
 |        V               V
 |     -------        --------
 |     \      \      /       /
 |      \      ------       /
 |       \       ALU       /
 |        -----------------
 |                |
 |                |
 +----------------+

ACCUMULATOR SOURCE      = REG 0
ACCUMULATOR DESTINATION = REG 0

ALU OPERATION BETWEEN TWO REGISTERS[edit | edit source]

   +-------------+        
   |    REG 0    |   FROM MEMORY 
   +-------------+	   |
          |               |
          |               | 
          V               V
       -------        --------
       \      \      /       /
        \      ------       /
         \       ALU       /
          -----------------
                  |
                  |
                  V        
           +-------------+
           |    REG 1    |
           +-------------+

ACCUMULATOR SOURCE      = REG 0
ACCUMULATOR DESTINATION = REG 1

INDEX REGISTERS[edit | edit source]

Index registers will have the same source and destination and are set with the two byte $42 (SIR) immediate opcode.

  • Immediate value bits [7:4] = Y Register
  • Immediate value bits [3:0] = X Register

SEGMENT MAPPER[edit | edit source]

The CPU can see 64k of contiguous memory. To access more than 64k the segment mapper sets the upper 8 bits of the CPUs 22 bit address bus. There are four 16k banks that may be set independently.

Setting banks can be achieved by loading or executing ALU operations with the accumulator, X or Y register destinations pointing to one of the four segment register files.

   +-------------+
   |    REG 12   |  BANK 0 SEGMENT
   | $0000-$3FFF | 
   |Default Value|
   |  %00000000  |  
   +-------------+
   |    REG 13   |  BANK 1 SEGMENT
   | $4000-$7FFF |
   |Default Value|  
   |  %00000001  |
   +-------------+
   |    REG 14   |  BANK 2 SEGMENT
   | $8000-$BFFF |
   |Default Value|  
   |  %00000010  |
   +-------------+
   |    REG 15   |  BANK 3 SEGMENT
   | $C000-$FFFF |
   |Default Value|  
   |  %00000011  |
   +-------------+

Bits[1:0] = AddressOut[15:14]
Bits[7:2] = AddressOut[21:16] 

   +-------------+
   |    REG 8    |  BANK 3 - 0  
   |Default Value|  Access Control
   |  %01010101  | 
   +-------------+

Bank 0 Bits[1:0] 
Bank 1 Bits[3:2] 
Bank 2 Bits[5:4]
Bank 3 Bits[7:6] 

          00 = ROM
          01 = RAM
          10 = Reserved
          11 = reserved

CPU CONTROL REG 9[edit | edit source]

  • Bit 0 Skip internal cycle when set
  • Bit 1 Burst enable when set

BRANCH ALWAYS[edit | edit source]

Branch always $12 (BRA) is a two byte relative opcode. BRA will branch relative 127 forward or 128 back.

OPTIMIZED MEMORY ACCESS[edit | edit source]

Memory accesses repeat on a 32-cycle pattern. All reads to SDRAM are performed in burst of 4 and writes are single access. SRAM, ROM and register writes are single read and single write.

CPU CYCLES[edit | edit source]

When skip internal cycles is set in the CPUs control register the instruction timing is as follows.

   0123456789ABCDEF
  +----------------
0x|66.7334422124455
1x|2526334414154455
2x|56.7334422124455
3x|2526334414154455
4x|4627334422123455
5x|25.6334414154455
6x|46.7334422125455
7x|25.6334414154455
8x|2626333312124444
9x|25.5333314144444
Ax|2626333312124444
Bx|25.5333314144444
Cx|2627334412124455
Dx|25.6334414154455
Ex|2627334412124455
Fx|25.6334414154455

When the burst bit is enabled the CPU will fetch 4 bytes at a time and will use them with instructions that have sequential memory accesses. For example immediate instructions have one opcode byte and one data byte in sequential order. You can execute 2 immediate instructions or 4 implied instructions per 1mhz cycle at max.

The CPU will halt burst execution any time there is a non-sequential read or any write.

The multi-byte burst fetches are on 4 byte boundaries. For maximum performance instructions that execute in sequential order (immediate, implied, absolute..) should start at word 0, so 4 bytes of data/instructions can be executed in the same time 1 instruction would be executed.

Examples:

All 4 instructions can execute before next memory access.

C000 LDA #$01
C002 ROR
C003 SEI

LDA can not execute in 1 memory access, since it crosses a 4-word boundary. Instructions C005-C007 all execute next memory access.

C003 LDA #$01
C005 ROR
C006 SEI
C007 NOP

Access 1 execution stops with reads to non-immediate memory locations. Memory access 2 will be from zero page and execution stops. Access 3 will execute instructions C002-C003.

C000 LDA $01
C002 ROR
C003 SEI

Access 1 execution stops with writes to non-immediate memory locations. Memory access 2 will be to zero page and execution stops. Access 3 will execute instructions C002-C003.

C000 STA $01
C002 ROR
C003 SEI

Access 1 executes all cycles, except the read to $D020. Access 2 reads from $D020 and cycle 3 executes C003.

C000 LDA $D020
C003 SEI

Cycle 1 executes up to the actual write. Cycle 2 does the write. Cycle 3+ executes your self-modified code. (Bad. Bad. Boo. Hiss.)

C000 STA $C003
C003 ...

Cycle 1 executes up to the actual read. Cycle 2 does the read. Cycle 3 gets stalled since STA's opcode isn't available (STA is not aligned). Cycle 4 reads the STA's opcode and executes up to the actual write. Cycle 5 does the write. Cycle 6 executes both NOPs.

C000 LDA $1000
C003 STA $4000
C006 NOP
C007 NOP

Cycle 1 executes up to the actual read. Cycle 2 does the read. Cycle 3 executes up to the actual write. Cycle 4 does the write.

C000 NOP
C001 LDA $1000
C004 NOP
C005 STA $4000

Cycle 1 executes up to the actual read. Cycle 2 does the read. Cycle 3 executes the NOP. Cycle 4 executes up to the actual write. Cycle 5 does the write. Cycle 6 executes the NOP.

C000 LDA $1000
C003 NOP
C004 STA $4000
C007 NOP

Cycle 1 executes up to the actual write to stack. Cycle 2 does the write. Cycle 3 executes NOPs.

C000 PHA
C001 NOP
C002 NOP
C003 NOP

Cycle 1 executes NOPs and SAC. Cycle 2 executes LDA and LDY. Cycle 3 executes RTS. 5006/5007 are never reached in skip+burst mode.

1000 NOP
1001 NOP
1002 SAC #$CC
1004 LDA #$01
1006 LDY #$00
5006 BRK
5007 NOP
5008 RTS

Misc[edit | edit source]

PALETTE[edit | edit source]

There are 16 adjustable colors. $0-f When chroma is set to 0 there is no modulation and can be used for white, black and grays.

DTV palette compatibility bit when set will distribute color 15 chroma across $10-$ff. This allows you to have 16 colors that can be changed with one write to a register.

Colors $0-$f are [chroma] [luma] from adjustable palette.

Colors $10-$ff are [chroma] [luma] from color decoder only.

Default PALLETTE
----------------
Color0Luma  = $0  black
Color1Luma  = $f  white
Color2Luma  = $6  Red
Color3Luma  = $e  cyan
Color4Luma  = $8  purple
Color5Luma  = $b  green
Color6Luma  = $6  blue
Color7Luma  = $f  yellow
Color8Luma  = $9  orange
Color9Luma  = $6  brown
Color10Luma = $b  light red
Color11Luma = $5  dark gray
Color12Luma = $7  medium gray
Color13Luma = $f  light green
Color14Luma = $a  light blue
Color15Luma = $a  light gray
 
Color0Chroma = $0 black
Color1Chroma = $0 white
Color2Chroma = $3 Red
Color3Chroma = $b cyan 
Color4Chroma = $5 purple
Color5Chroma = $d green
Color6Chroma  = $8  blue
Color7Chroma  = $f  yellow
Color8Chroma  = $2  orange
Color9Chroma  = $2  brown
Color10Chroma = $3  light red
Color11Chroma = $0  dark gray
Color12Chroma = $0  medium gray
Color13Chroma = $d  light green
Color14Chroma = $9  light blue
Color15Chroma = $0  light gray

Default VIC Registers[edit | edit source]

raster_compare <= (others => '1')
light_pen_irq_en	<= '0'
sprite_sprite_irq_en<= '0';
sprite_background_irq_en	<= '0';
raster_irq_en		<= '0';
ExtendedRegEnableB		<= '0';
GBankA		<= (others => '0');
GBankB		<= (others => '0');
LinearAddressing	<= '0';
HiColor		<= '0';
ColorBankHigh	<= "0000";	
ColorBankLow	<= "01110110";
border_color	<= (others => '0');
bkgnd0_color	<= (others => '0');
bkgnd1_color	<= (others => '0');
bkgnd2_color	<= (others => '0');
bkgnd3_color	<= (others => '0');
ExtendedRegKill	<= '0'; 
PAL		<= '0';
bmm		<= '0';
ecm		<= '0';
vm		<= (others => '0');
cb		<= (others => '0');
AlwaysSetToZero	<= '0';
c_sel		<= '0';
r_sel		<= '0';
x_scroll		<= (others => '0');
y_scroll		<= (others => '0');
den		<= '0';
clear_light_pen_irq	<= '0';
clear_sprite_sprite_irq	<= '0';
clear_sprite_background_irq	<= '0';
clear_raster_irq		<= '0';
Sprite7Priority		<= '0';
Sprite6Priority		<= '0';
Sprite5Priority		<= '0';
Sprite4Priority		<= '0';
Sprite3Priority		<= '0';
Sprite2Priority		<= '0';
Sprite1Priority		<= '0';
Sprite0Priority		<= '0';
sprite_colora	<= (others => '0');
sprite_colorb	<= (others => '0');
sprite_0_color	<= (others => '0');
sprite_1_color	<= (others => '0');
sprite_2_color	<= (others => '0');
sprite_3_color	<= (others => '0');
sprite_4_color	<= (others => '0');
sprite_5_color	<= (others => '0');
sprite_6_color	<= (others => '0');
sprite_7_color	<= (others => '0');
PhaseAlternate	<= '0';		
BurstRate	<= "000111000001001000101010"; --new 18
OldDTVCompatibility	<= '0';
BorderOff		<= '0';
IrqTriggerCycle	<= "1000000"; --at the end of cycle 64
SpriteBank	<= (others => '0')
LinearModuloA	<= (others => '0')
LinearStartA	<= "0000000000000001110110"; --color bank LinearStepA	<= (others => '0')
LinearModuloB	<= (others => '0')
LinearStartB	<= (others => '0')
LinearStepB	<= (others => '0')
OverScan		<= '0'
ColorDisable	<= '0'
CPUBadlineDisable	<= '0'
ChunkyEnable	<= '0'
LineAdjust	<= "00001101"; 
PhaseAdjust	<= (others => '0')
mcm		<= '0'

MODULO PROGRAMMING[edit | edit source]

The DTV contains 3 locations where address are calculated with modulus counters. VIC, DMA and Blitter.

Modulus counters can be used to format data fetches from a contiguous memory.

Start Address     End of line counter
|                 cleared and modulus
|                 added to address 
|                 counter
|                            |
|                            |
+> +---------------------+ <-+
   |                     | <-+
   |                     | <-+
   |      Display        |  ...
   |       Window        |
   |                     |
   |                     |   
   +---------------------+

Displaying or moving portions of images that are bigger than the display window can be achieved by properly setting the modulus values.

Example:

  • 320x200 display window
  • 800x400 image

Set line count to 320, which is the number of pixels in a scan line.

Set modulus to 800(total pixels in image) - 320(scan line pixels)

Start address at beginning of image.

+-----------------+--------------+
|                 |              |
| Display window  |              |
|                 |              |
|                 |              |
|                 |              |
+-----------------+              |
|                                |
|                                |
+--------------------------------+

Moving start address along the first line will scroll horizontally in the image.

+---+----------------+-----------+
|   |                |           |
|   | Display window |           |
|   |                |           |
|   |                |           |
|   |                |           |
|   +----------------+           |
|                                |
|                                |
+--------------------------------+

Moving start by image horizontal size (In this example 800) will scroll the image down by 1 line.

+--------------------------------+
+-----------------+              |
|                 |              |
| Display window  |              |
|                 |              |
|                 |              |
|                 |              |
+-----------------+              |
|                                |
|                                |
+--------------------------------+

Unlimited scrolling in 640 x 400 buffer. Scroll = 0,0

+----------------+--------------+
|                |              |
|                |              |
|                |              |
+----------------+              |
|                               |
|                               |
|                               | 
+-------------------------------+

Scrolled down to 0,7 5120 bytes of new data plotted by blitter, placed at bottom of view port and exact copy placed above view port.

+-------------------------------+
|1111111111111111               |
+----------------+              |
|                |              |
|                |              |
|1111111111111111|              |
+----------------+              |
|                               |
|                               |
|                               | 
+-------------------------------+

Scrolled down to 0,319 5120 bytes of new data plotted by blitter, placed at bottom of view port and exact copy placed above view port.

+-------------------------------+
|1111111111111111               |
|    . . .                      |
|    . . .                      |
|NNNNNNNNNNNNNNNN               |
+----------------+              |
|    . . .       |              |
|    . . .       |              |
|NNNNNNNNNNNNNNNN|              | 
+----------------+--------------+

If y position of view port = 320 then next scroll will jump back to y position = 0. This allows unlimited y scrolling.

The same concept works in the x-axis and will jump to x position = 0 when x position = 640.

Combining both x and y are possible allowing scrolling in any direction with only 8320 bytes of blits per 8 pixel plotting.

Moving by multiples of lines and pixels can be used to scroll in all directions.

VIC Start addresses are loaded on line 49 and line 11 in over-scan mode.

Hardware issues (for the DTV1)[edit | edit source]

May be outdated. Some might still apply for the DTV2/DTV3.

  • nDMAWrite signals an external write during DMA DMACycles. nIORD and nIOWR are output only.
  • USR[7:0] port is output only due to a synthesis error.
  • Pixel[7:0] are raw pixel data output before luma and chroma lookup tables.
  • PS-2 data and clk are for keyboards only. Keyboard statemachine has issues with f7 and pressing shift at the same time as another key.
  • nNMI may have slight timing problems that cause digi playback problems.
  • Cartridge port signals are generated for internal use. There may be inversion or timing delay problems. (Better check before using)
  • IO's on the ASIC are not 5v tolerant, so put 100ohm series resistors.
    • Check this - doesn't apply at least to PS/2 and IEC lines
  • Expanded sprites have 1/4 pixel gap and may have a problem with y expansion in the left border.
  • VIC has small timing delays with color update times, which causes problems with disk throw in SummerGames.

C64 emulation issues (for the DTV2/3)[edit | edit source]

  • CPU
    • PLP/PHP: sometimes bit 5 in the CPU status register is unset
  • CIA
    • TOD (time of day) timer is missing ($DC08-$DC0B/$DD08-$DD0B)
    • Keyboard emulation is flawed (writing to $DC00 and reading from $DC01 works but not the other way around)
    • The original CIA has a one-cycle latency between writing to the register and outputting the data on its port pins. The one in the DTV is faster, leading to problems with certain fastloaders (PETSCII posting)
  • SID
  • NMI timing is not accurate
  • Some illegal OPs are missing, some guesses at what they do instead:
    • SHS ($9b) is a nop absolute,y
    • DCP ($c7) is zeropage, read modify write, decrement and compare against the accumulator
    • ANC ($2b) it a 2 byte nop
    • ($fc) is a nop absolute,x
  • VICII
    • idle fetch is different (see _draw_idle in VICEplus vicii-draw.c), leading to deviant behaviour in screen border
    • DMA delay has a fixed delay of one, breaking Variable Screen Positioning (VSP)
    • $D019 bit 7 is (bit 0 | bit 1 | bit 2 | bit 3) and gets set even if these conditions were not enabled in $D01A (and, thus, didn't actually trigger an IRQ)
    • Reading $D01E/$D01F does not clear the corresponding $D019 bits - in contrast to the C64, these are sticky and have to get ACK'd explicitly.
  • Misc Hardware Issues
    • PAL output signal is known not to work with some digital displays (flat panel displays, FBAS=>VGA converters, TV cards, ...). Setting $D04E to 2 or 3 seems to help sometimes (PETSCII thread)
    • PS/2 keyboard acts strange after powerup (after a reset everything is okay though). F7 key does not work; shift keys are swapped. See here for possible fixes

For details, see VICEplus x64dtv emulator pages.

Hardware specifications[edit | edit source]

RATINGS[edit | edit source]

VDD                        3.3v
MAX                        3.6v
Operating Temp             0C - 70C
High Level Input           1.7v
Low Level Input            1.1v
Schmitt hysteresis          .6v
Capacitance Input (die)    2.4pF
Capacitance Output (die)   5.6pF
Capacitance Bidir (die)    6.6pF 

PIN ASSIGNMENTS[edit | edit source]

Input Pins            
-----------------+-----
Name             ;  
-----------------+-----

ATNIn        ; 5v Tol 
Clk32mhz     ; 5v Tol 
KeyboardClk  ; 5v Tol
KeyboardData ; 5v Tol
LightPen     ; 5v Tol ;Schmitt
nDMA         ; 5v Tol
nReset       ; 5v Tol
nSRAMSelect  ; 5v Tol
nSVideo      ; 5v Tol

Output Pins            
-----------------+-----
Name             ;  
-----------------+-----
AddressBufferDir ; LVTTL 
CPUAddressEn     ; LVTTL  
CSync            ; LVTTL  
Chroma[0]        ; LVTTL ; 12ma  
Chroma[1]        ; LVTTL ; 12ma 
Chroma[2]        ; LVTTL ; 12ma
Chroma[3]        ; LVTTL ; 12ma 
Clk1mhzEn        ; LVTTL  
DataBufferDir    ; LVTTL  
DataBuffer_nOE   ; LVTTL   
IECATN           ; LVTTL  
Luma[0]          ; LVTTL ; 12ma
Luma[1]          ; LVTTL ; 12ma
Luma[2]          ; LVTTL ; 12ma
Luma[3]          ; LVTTL ; 12ma
SDRAMCLK         ; LVTTL ; Low Skew 
SDRAMLDM         ; LVTTL   
SDRAMUDM         ; LVTTL   
SDRAM_nCS        ; LVTTL   
SDRAMnCAS        ; LVTTL   
SDRAMnRAS        ; LVTTL   
Voice1Sigma      ; LVTTL ; 12ma
Voice2Sigma      ; LVTTL ; 12ma
Voice3Sigma      ; LVTTL ; 12ma
VolumeSigma      ; LVTTL ; 12ma
nIORd            ; LVTTL 
nRAMCS           ; LVTTL 
nROMCs           ; LVTTL 
nWrite           ; LVTTL   
-----------------+-----

-----------------------
Bidir Pins             
------------+-------+--
Name        ; Pin # ; I
------------+-------+--
Address[0]  ; LVTTL ; 5V Tol
Address[10] ; LVTTL ; 5V Tol
Address[11] ; LVTTL ; 5V Tol
Address[12] ; LVTTL ; 5V Tol
Address[13] ; LVTTL ; 5V Tol 
Address[14] ; LVTTL ; 5V Tol
Address[15] ; LVTTL ; 5V Tol
Address[16] ; LVTTL ; 5V Tol 
Address[17] ; LVTTL ; 5V Tol
Address[18] ; LVTTL ; 5V Tol
Address[19] ; LVTTL ; 5V Tol
Address[1]  ; LVTTL ; 5V Tol 
Address[20] ; LVTTL ; 5V Tol
Address[21] ; LVTTL ; 5V Tol
Address[2]  ; LVTTL ; 5V Tol
Address[3]  ; LVTTL ; 5V Tol
Address[4]  ; LVTTL ; 5V Tol
Address[5]  ; LVTTL ; 5V Tol
Address[6]  ; LVTTL ; 5V Tol
Address[7]  ; LVTTL ; 5V Tol
Address[8]  ; LVTTL ; 5V Tol
Address[9]  ; LVTTL ; 5V Tol
Data[0]     ; LVTTL ; 5V Tol
Data[1]     ; LVTTL ; 5V Tol
Data[2]     ; LVTTL ; 5V Tol
Data[3]     ; LVTTL ; 5V Tol
Data[4]     ; LVTTL ; 5V Tol
Data[5]     ; LVTTL ; 5V Tol
Data[6]     ; LVTTL ; 5V Tol
Data[7]     ; LVTTL ; 5V Tol
IECClk      ; LVTTL ; 5V Tol ;Pullup
IECData     ; LVTTL ; 5V Tol ;Pullup 
JoyA[0]     ; LVTTL ; 5V Tol ;Pullup
JoyA[1]     ; LVTTL ; 5V Tol ;Pullup
JoyA[2]     ; LVTTL ; 5V Tol ;Pullup
JoyA[3]     ; LVTTL ; 5V Tol ;Pullup
JoyA[4]     ; LVTTL ; 5V Tol ;Pullup
JoyA[5]     ; LVTTL ; 5V Tol ;Pullup
JoyB[0]     ; LVTTL ; 5V Tol ;Pullup
JoyB[1]     ; LVTTL ; 5V Tol ;Pullup
JoyB[2]     ; LVTTL ; 5V Tol ;Pullup
JoyB[3]     ; LVTTL ; 5V Tol ;Pullup
JoyB[4]     ; LVTTL ; 5V Tol ;Pullup
JoyB[5]     ; LVTTL ; 5V Tol ;Pullup
PA2         ; LVTTL ; 5V Tol ;Pullup
Paddle      ; LVTTL ; 5V Tol ;Schmitt
USR[0]      ; LVTTL ; 5V Tol ;Pullup
USR[1]      ; LVTTL ; 5V Tol ;Pullup
USR[2]      ; LVTTL ; 5V Tol ;Pullup
USR[3]      ; LVTTL ; 5V Tol ;Pullup
USR[4]      ; LVTTL ; 5V Tol ;Pullup
USR[5]      ; LVTTL ; 5V Tol ;Pullup
USR[6]      ; LVTTL ; 5V Tol ;Pullup
USR[7]      ; LVTTL ; 5V Tol ;Pullup
nIRQ        ; LVTTL ; 5V Tol ;Pullup
nNMI        ; LVTTL ; 5V Tol ;Pullup

Input Pins            
-----------------+-----
Name             ;  
-----------------+-----

CPUDataPort[4]        
  Input port to bit 4 of register $0000 and $0001 

Clk32mhz
  ~32Mhz clock input (PAL 31.36, NTSC 32.64)
  
KeyboardClk  
  PS/2 Keyboard clock

KeyboardData 
  PS/2 Keyboard Data 

LightPen
  Active low lightpen trigger
     
nDMA         
  External DMA (active low).  Tristates data, address and nWrite.

nReset       
  Global reset(active low) synchronized to system clock.

nSRAMSelect  
  Disables SDRAM when low and nRAMCS activated.

nSVideo      
  Selects Separate luma and chroma when low.  (internally mixed when high)

Output Pins            
-----------------+-----
Name             ;  
-----------------+-----

CPUAddressEn      
  Indicates CPU address cycle when high.

CSync             
  Drives high during non sync times to set black level.

Chroma[0]         
Chroma[1]        
Chroma[2]        
Chroma[3]        
  Chrominance output during nSVIDEO = gnd, otherwise composite.
 
Clk1mhzEn        
  32mhz strobe at CPU execution.
 
DataBufferDir 
  Controls direction of buffers if long external bus used.  
   
DataBuffer_nOE   
  Controls output of buffers if long external bus used.  
   
IECATN           
  Disk serial attention.
  
Luma[0]          
Luma[1]          
Luma[2]          
Luma[3]
  Luminace output during nSVIDEO = gnd, otherwise composite.
          
SDRAMCLK             
SDRAMLDM            
SDRAMUDM            
SDRAM_nCS           
SDRAMnCAS           
SDRAMnRAS        
  SDRAM control signals.
   
Voice1Sigma      
Voice2Sigma      
Voice3Sigma      
VolumeSigma      
  Sigma converts.  Must be run through a lowpass filter.

nIORd            
  Active low read
 
nRAMCS           
  Active low SRAM chips elect when nSRAM = gnd

nROMCs            
  Active low ROM chip select.

nWrite              
  Global write for SRAM, Flash and SDRAM.
  This should be buffered if used on long external bus

-----------------------
Bidir Pins             
------------+-------+--
Name        ; Pin # ; I
------------+-------+--
Address[0]   
Address[10] 
Address[11]
Address[12]
Address[13]
Address[14]  
Address[15] 
Address[16] 
Address[17] 
Address[18] 
Address[19] 
Address[1]  
Address[20] 
Address[21] 
Address[2]  
Address[3]  
Address[4]  
Address[5]  
Address[6]  
Address[7]  
Address[8]  
Address[9]
  Address should be buffered if used on long external bus. 
 
Data[0]     
Data[1]     
Data[2]     
Data[3]     
Data[4]     
Data[5]     
Data[6]     
Data[7]     
  Data should be buffered if used on long external bus.

IECClk
  Disk clock.
      
IECData
  Disk Data.
     
JoyA[0]     
JoyA[1]     
JoyA[2]     
JoyA[3]     
JoyA[4]     
JoyA[5]     
JoyB[0]     
JoyB[1]     
JoyB[2]     
JoyB[3]     
JoyB[4]     
JoyB[5]     
  Open collector joystick ports.

PA2         
  CIA PA line.

Paddle      
  Charge dump analog A/D converter.

USR[0]      
USR[1]      
USR[2]      
USR[3]      
USR[4]      
USR[5]      
USR[6]      
USR[7]      
  Open collector user port pins

nIRQ        
  Negative assert IRQ (bidir!)

nNMI        
  Negative assert NMI (bidir!)


Pin Locations
---------------------------------------
160 <CORNER>
159 VSS
158 VDD
157 VSS
156 VDD
155 TMODE
154 USR_0
153 USR_1
152 VSS
151 VDD
150 USR_2
149 USR_3
148 USR_4
147 USR_5
146 VSS
145 VDD
144 USR_6
143 USR_7
142 Luma_0
141 Luma_1
140 VSS
139 VDD
138 Luma_2
137 Luma_3
136 JoyB_0
135 JoyB_1
134 VSS
133 VDD
132 JoyB_2
131 JoyB_3
130 JoyB_4
129 JoyB_5
128 VSS
127 VDD
126 JoyA_0
125 JoyA_1
124 JoyA_2
123 JoyA_3
122 VSS
121 <CORNER>
120 <CORNER>
119 VDD
118 VSS
117 VDD
116 JoyA_4
115 JoyA_5
114 Data_0
113 Data_1
112 VSS
111 VDD
110 Data_2
109 Data_3
108 Data_4
107 Data_5
106 VSS
105 VDD
104 Data_6
103 Data_7
102 Chroma_0
101 Chroma_1
100 VSS
99  VDD
98  Chroma_2
97  Chroma_3
96  Address_0
95  Address_1
94  VSS
93  VDD
92  Address_2
91  Address_3
89  Address_5
88  VSS
87  VDD
86  Address_6
85  Address_7
84  Address_8
83  Address_9
82  VSS
81  <CORNER>
80  <CORNER>
79  VDD
78  VSS
77  VDD
76  Address_10
75  Address_11
74  Address_12
73  Address_13
72  VSS
71  VDD
70  Address_14
69  Address_15
68  Address_16
67  Address_17
66  VSS
65  VDD
64  Address_18
63  Address_19
62  Address_20
61  Address_21
60  VSS
59  VDD
58  Paddle
57  PA2
56  nNMI
55  nIRQ
54  VSS
53  VDD
52  IECData
51  IECClk
50  Clk1mhzEn
49  CPUAddressEn
48  VSS
47  VDD
46  VolumeSigma
45  Voice3Sigma
44  Voice2Sigma
43  Voice1Sigma
42  VSS
41  <CORNER>
40  <CORNER>
39  VDD
38  VSS
37  VDD
36  CSync
35  nRAMCS
34  nROMCs
33  nWrite
32  VSS
31  VDD
30  nIORd
29  SDRAMnRAS
28  SDRAMnCAS
27  IECATN
26  VSS
25  VDD
24  SDRAMLDM
23  SDRAMUDM
22  SDRAM_nCS
21  DataBuffer_nOE
20  VSS
19  VDD
18  DataBufferDir
17  AddressBufferDir
16  clko
15  Clk32mhz
14  SDRAMCLK
13  VSS
12  VDD
11  nSVideo
10  nSRAMSelect
9   LightPen
8   ATNIn
7   VSS
6   VDD
5   nDMA
4   nReset
3   KeyboardData
2   KeyboardClk
1  <CORNER>

PAD COORDINATES[edit | edit source]

Please see the die pad coordinates below. They're all measured from the center of the die, so X coordinates are negative to the left of center and positive to the right, while Y coordinates are negative below center and positive above.

Also, the coordinates are scaled by 10 for some reason, including the die size, itself, which should be 3.785mm x 3.759mm. Die pad

  1. 1 is given as "-15.109 17.335", but this translates to X being

1.5109mm left of center and Y being 1.7335mm above center.

#Pads list
1 -15.109 17.335
2 -14.311 17.335
3 -13.513 17.335
4 -12.715 17.335
5 -11.917 17.335
6 -11.119 17.335
7 -10.321 17.335
8 -9.523 17.335
9 -8.725 17.335
10 -7.927 17.335
11 -7.129 17.335
12 -6.331 17.335
13 -5.533 17.335
14 -4.735 17.335
15 -3.937 17.335
16 -3.139 17.335
17 -2.341 17.335
18 -1.543 17.335
19 -0.745 17.335
20 0.053 17.335
21 0.851 17.335
22 1.649 17.335
23 2.447 17.335
24 3.245 17.335
25 4.043 17.335
26 4.841 17.335
27 5.639 17.335
28 6.437 17.335
29 7.235 17.335
30 8.033 17.335
31 8.831 17.335
32 9.629 17.335
33 10.427 17.335
34 11.225 17.335
35 12.023 17.335
36 12.821 17.335
37 13.619 17.335
38 14.417 17.335
39 15.215 17.335
40 16.013 17.335
41 17.335 15.095
42 17.335 14.297
43 17.335 13.499
44 17.335 12.701
45 17.335 11.903
46 17.335 11.105
47 17.335 10.307
48 17.335 9.509
49 17.335 8.711
50 17.335 7.913
51 17.335 7.115
52 17.335 6.317
53 17.335 5.519
54 17.335 4.721
55 17.335 3.923
56 17.335 3.125
57 17.335 2.327
58 17.335 1.529
59 17.335 0.731
60 17.335 -0.066
61 17.335 -0.864
62 17.335 -1.662
63 17.335 -2.460
64 17.335 -3.258
65 17.335 -4.056
66 17.335 -4.854
67 17.335 -5.652
68 17.335 -6.450
69 17.335 -7.248
70 17.335 -8.046
71 17.335 -8.844
72 17.335 -9.642
73 17.335 -10.440
74 17.335 -11.238
75 17.335 -12.036
76 17.335 -12.834
77 17.335 -13.632
78 17.335 -14.430
79 17.335 -15.228
80 17.335 -16.026
81 15.095 -17.335
82 14.297 -17.335
83 13.499 -17.335
84 12.701 -17.335
85 11.903 -17.335
86 11.105 -17.335
87 10.307 -17.335
88 9.509 -17.335
89 8.711 -17.335
90 7.913 -17.335
91 7.115 -17.335
92 6.317 -17.335
93 5.519 -17.335
94 4.721 -17.335
95 3.923 -17.335
96 3.125 -17.335
97 2.327 -17.335
98 1.529 -17.335
99 0.731 -17.335
100 -0.066 -17.335
101 -0.864 -17.335
102 -1.662 -17.335
103 -2.460 -17.335
104 -3.258 -17.335
105 -4.056 -17.335
106 -4.854 -17.335
107 -5.652 -17.335
108 -6.450 -17.335
109 -7.248 -17.335
110 -8.046 -17.335
111 -8.844 -17.335
112 -9.642 -17.335
113 -10.440 -17.335
114 -11.238 -17.335
115 -12.036 -17.335
116 -12.834 -17.335
117 -13.632 -17.335
118 -14.430 -17.335
119 -15.228 -17.335
120 -16.026 -17.335
121 -17.335 -15.109
122 -17.335 -14.311
123 -17.335 -13.513
124 -17.335 -12.715
125 -17.335 -11.917
126 -17.335 -11.119
127 -17.335 -10.321
128 -17.335 -9.523
129 -17.335 -8.725
130 -17.335 -7.927
131 -17.335 -7.129
132 -17.335 -6.331
133 -17.335 -5.533
134 -17.335 -4.735
135 -17.335 -3.937
136 -17.335 -3.139
137 -17.335 -2.341
138 -17.335 -1.543
139 -17.335 -0.745
140 -17.335 0.053
141 -17.335 0.851
142 -17.335 1.649
143 -17.335 2.447
144 -17.335 3.245
145 -17.335 4.043
146 -17.335 4.841
147 -17.335 5.639
148 -17.335 6.437
149 -17.335 7.235
150 -17.335 8.033
151 -17.335 8.831
152 -17.335 9.629
153 -17.335 10.427
154 -17.335 11.225
155 -17.335 12.023
156 -17.335 12.821
157 -17.335 13.619
158 -17.335 14.417
159 -17.335 15.215
160 -17.335 16.013

SDRAM Row/Column Address mapping[edit | edit source]

Address bus during row accesses 
(Address[15:12], Address[19:16], Address[15:8])

Address bus during column accesses 
(Address[15:12], Address[19],'1',"00", Address[7:0])

SDRAM Read
---------------------------------------
Clock Cycle 0  1  2  3  4  5  6  7
SDRAMDM    ------\_________/-------
SDRAM_nCS  ---\____/---------------
SDRAMnRAS  ---\_/------------------
SDRAMnCAS  ------\_/---------------
nWrite     ------------------------
DataIn                  0  1  2  3 
Address      Row/Col<-Flat Static->


SDRAM Write 
---------------------------------------
Clock Cycle 0  1  2  3  4  5  6  7
SDRAMDM    ------\_/---------------
SDRAM_nCS  ---\____/---------------
SDRAMnRAS  ---\_/------------------
SDRAMnCAS  ------\_/---------------
nWrite     ------\_/---\_______/---
DataOut           XXXXXXXXXXXXXXXXX
Address      Row/Col<-Flat Static->

SRAM/ROM Read
---------------------------------------
Clock Cycle 0  1  2  3  4  5  6  7
CPUAddrEn  /----------------------\_
nROMCS     ---------\_____________/-
nIORD      ------------\__________/-
nWrite     -------------------------
DataIn                           X 
Address      Row/Col<-Flat Static->

SRAM/Flash Write
---------------------------------------
Clock Cycle 0  1  2  3  4  5  6  7
CPUAddrEn  /----------------------\_
nROMCS     ---------\_____________/-
nIORD      -------------------------
nWrite     ------\_/---\_______/----
DataOut           XXXXXXXXXXXXXXXXX
Address      Row/Col<-Flat Static->

Clk1mhzEn during CPU access.
---------------------------------------
Clock Cycle 0  1  2  3  4  5  6  7
Clk1mhzEn   ____________________/-\_


32 cycle accesses(1mhz)
---------------------------------------
Clock Cycle<0-7> <8-15> <16-23> <24-31>
            Char  Color   GFX    CPU
             or     or     or      
            PlaneA DMA    PlaneB   
                    or                                  
                   Blit

Clk1mhzEn strobes during cycle 31.

A note on global set and reset[edit | edit source]

When TMODE is high ; the chip is in "test" mode and now nNMI and nIRQ are active high global set and active high global reset pins respectively.

Toggling these 2 pins when TMODE is high will put all the flops in a known and defined state.

Startup User Port Bits[edit | edit source]

On power the user port is polled by the boot ROM and video modes are set by the state the bits.

0      PAL/NTSC Line timing (1=PAL 
       63cycles/line)
1      PAL/NTSC burst alternation 
       enable (1=alternate)
2      Saturation 0
3      Saturation 1
4      Burst lock enable (1=enable)
5      Burst lock type
6      Line timing fine tune (1=PAL)
7      PAL/NTSC burst select (1=PAL)
D6510bit[4] 1=old xtals, 0=new

Coding Examples: Multiply[edit | edit source]

Original[edit | edit source]

;8 bit multiply with 16 bit product
; MULND(8) * MULR(8) = PROD(16)
 
MULTIPLY:  lda     #$00   ;Clear lower half of product
           sta     PROD+1 ;Clear upper half of product
           ldx     #8     ;Set count
SHIFT:     asl     a      ;Shift product left one bit
           rol     PROD+1
           asl     MULR   ;Shift multiplier left
           bcc     CHCNT  ;No addition if next bit is zero
           clc
           adc     MULND
           bcc     CHCNT
           inc     PROD
CHCNT      dex
           bne     SHIFT
           sta     PROD

Using register bank[edit | edit source]

; macros need to be fancier...
.MACRO      AS0D0  .BYT $32, %00000000
.MACRO      AS3D3  .BYT $32, %00110011
.MACRO      AS4D4  .BYT $32, %01000100

MULTIPLY:  lda     #$00   ;Clear lower half of product
           AS3D3          ;Use reg 3 as storage
           lda     #$00   ;Clear upper half of product
           AS4D4          ;Use reg 4 as storage
           lda     MULR
           ldx     #8     ;Set count
SHIFT:     AS0D0          ;Back to reg 0
           asl     a      ;Shift product left one bit
           AS3D3
           rol
           AS4D4 
           asl            ;Shift multiplier left
           bcc     CHCNT  ;No addition if next bit is zero
           clc
           adc     MULND  ;Can't use extra reg's for adds :(
           bcc     CHCNT
           AS3D3
           clc
           adc     #1     ;This is an "inc a" - should have added that op code like the c02 has :)
CHCNT      dex
           bne     SHIFT
           AS3D3
           sta     PROD+1
           AS0D0
           sta     PROD

This can be optimized more by - pointing the accumulator to the y register so you can iny the accumulator, by pre-decrementing the loop count, and by self modifying the adc multnd (changing to an immediate add) which is *naughty*.

A bit faster[edit | edit source]

...using Y high-byte of a 16bit A. (also, last sample had a bug)

; macros need to be fancier...
.MACRO AS0D0 .BYT $32, %00000000
.MACRO AS2D2 .BYT $32, %00100010
; a is now y
.MACRO AS3D3 .BYT $32, %00110011
MULTIPLY: lda #$00   ;Clear lower half of product
          tay        ;Clear upper half of product
          AS3D3      ;Use reg 3 as storage
          lda MULR
          AS0D0      ;Back to reg 0
          ldx #8     ;Set count
 SHIFT:   asl a      ;Shift product left one bit
          AS2D2
          rol
          AS3D3
          asl        ;Shift multiplier left
          bcc CHCNT  ;No addition if next bit is zero
          AS0D0
          clc
          adc MULND  ;Can't use extra reg's for adds :(
          bcc CHCNT
          iny        ;This is like an "inc a"
          CHCNT
          dex
          bne SHIFT
          sty PROD+1
          AS0D0
          sta PROD

Faster yet by re-doing the loop structure.[edit | edit source]

; macros need to be fancier...
.MACRO AS0D0 .BYT $32, %00000000
.MACRO AS2D2 .BYT $32, %00100010
; a is now y
.MACRO AS3D3 .BYT $32, %00110011
MULTIPLY: lda #$00    ; Clear lower half of product
          tay         ; Clear upper half of product
          AS3D3       ; Use reg 3 as storage
          lda MULR
          AS0D0 0     ; Back to reg 0
          ldx  #9     ; Set count
CHCNT:    dex
          beq DONE
SHIFT:    asl a       ; Shift product left one bit
          AS2D2
          rol
          AS3D3
          asl         ; Shift multiplier left
          bcc CHCNT   ; No addition if next bit is zero
          AS0D0
          clc
          adc MULND   ;Can't use extra reg's for adds :(
          bcc CHCNT
          iny         ;This is like an "inc a"
          bcs CHCNT   ;Could use new BRA op code
DONE:     sta PROD
          sty PROD+1

Self-modifying code[edit | edit source]

Hey! Let's go a bit faster by being "bad"... oh, so very bad!!

; macros need to be fancier...
.MACRO AS0D0 .BYT $32, %00000000
.MACRO AS2D2 .BYT $32, %00100010
; a is now y
.MACRO AS3D3 .BYT $32, %00110011
MULTIPLY: lda MULND
          sta BADME+1
          lda #$00     ;Clear lower half of product
          tay          ;Clear upper half of product
          AS3D3        ;Use reg 3 as storage
          lda MULR
          AS0D0        ;Back to reg 0
          ldx #9       ;Set count
CHCNT:    dex
          beq DONE
SHIFT:    asl a        ;Shift product left one bit
          AS2D2
          rol
          AS3D3
          asl          ;Shift multiplier left
          bcc CHCNT    ;No addition if next bit is zero
          AS0D0
          clc
BADME     adc #0       ;Oh so bad, self-modifying code!!!!!!!
          bcc CHCNT
          iny          ;This is like an "inc a"
          bcs CHCNT    ;Could use new BRA op code
DONE:     sta PROD
          sty PROD+1

With this, there is no burst fetch break when getting data, it only breaks on the 2 branches it takes in the loop. Putting SHIFT and CHCNT on word boundaries might make things a bit faster yet.