Learn Multi platform Z80 Assembly Programming... With Vampires!
Platform Specific Lessons
<- Back to the Main Contents & Basic Z80 Assembly Lessons





Lesson P60 - Tape loading on the Amstrad CPC (5K subs special)
The 5000 sub special demo used a cassette file, and streamed data in a couple of formats to the screen.

Lets take a look at the loading methods used.

CPC_TapeRLE.asm
5K_Demo.cdt


Making a tape with 2CDT
We're using 2CDT to build a tape file... Our test program has two types:
2 files are saved normally with blocks (The RLE compressed Cinema)
1 file is saved headerless (The final thanks message)

2CDT takes a variety of switches we'll need:

;-n   = start New tape
;-s 1 = Speed 2000 Baud (fast)
;-m 0 = Add Block data (use with &BC77)
;-t 1 = Turbo Loading blocks (default)

;-m 1 = Add Headerless (use with &BCA1)

Tape loading... Getting Started
Our demo is started with CALL &4000 ...

We're running on an AMSTRAD 6128 - this system starts with reading from disk - we need to use the basic command "|TAPE"
There's no defined firmware command to do this, but we can do a 'FAR CALL' (Call to ROM) with RST3

This command takes a pointer to 3 bytes... the first two are a destination address (&CCFD for |TAPE) and the third is a ROM NUMBER (7 for the DISK ROM which contains the command)



By default the CPC will show tape messages - we'll want to turn this off.
We do this with &BC6B (CAS NOISY)... passing this a 0 in A will enable messages (Default)... we want to turn them off so we set A=1

Tape loading... Reading Bytes (With Blocks)

We're going to open a file on the tape... it's RLE data, and we'll process it a byte at a time.

We need to specify a memory address of a 2k buffer - we do this with DE

If we want to specify the filename to load, we specify a pointer to the name in HL - and the filename length in B... alternatively we can specify B=0 to just load the next file on the tape.

We then use function &BC77 (CAS IN OPEN) to open the cassette file
we use &BC80 (CAS IN CHAR) to read in a single byte from the file...

We're using the RLE decoder we looked at previously to show the cinema graphic

Each section has a 'header byte'
If the whole byte is zero - then we're reached the end of the compressed sequence
If the top bit is 1, the next sequence is RLE compressed... the remaining bits are a bytecount (1-127)
If the top bit is 0, the next sequence is LINEAR uncompressed data... the remaining bits are a bytecount (1-127)

The RLE decompressor has been modified to skip zero bytes, allowing multiple layers to be overlaid to build up an image.

Layer 1
Layer 2
Combined Result
Once we've loaded in the file we use &BC7A (CAS IN CLOSE) to close the file.
If you don't want to do anything clever with the file byte by byte, and want to read in blocks, use &BC83 - CAS IN DIRECT

Specify a destination address with HL,and the file will be loaded into ram.

Tape loading... Reading to ram (Headerless)

If we want to quickly load a large block of data, Headerless data is fastest... we load headerless data with &BCA1 (CAS READ)

As there's no header, we don't specify a filename... there's no blocks, so the data is quicker.

We need to specify a destination in HL, a size in DE... and a Header sync byte in A (this needs to match the one used to build the file - 2CDT and the amstrad BIOS default to &16 for data) 
This can be used for reading in a loading screen in one go.




Lesson P61 - 2x and 4x sprite scaling with Lookup Tables on the Amstrad CPC

Lets take a look at sprite scaling! in this example we'll use lookup tables to effect Double and Quad size sprites on the Amstrad CPC.

We'll use a lookup table, so the scaled sprite will be about as fast as a regular one.

CPC_Scale2x4x_Sprite.asm

This example scales images on the CPC... but the principal will work on any system which uses a bitmap screen (Like the SAM or Speccy)

Scaling sprites can be used for a neat effect - it could also be used to save memory if you want huge graphics - of course a sprite scaled 2x in both directions uses 1/4 of the size of an unscaled sprite.

Sprite scaling and the Lookup tables

We're going to scale our Chibiko sprite - here it's shown at 1x , 2x and 4x size.

The sprite file is the same as the one used in the simple series. and the code used to display the 1x version is almost the same (It uses different registers for convenience to match the 2x and 4x versions)
We need some lookup tables to convert the bytes to 2x and 4x... these contain the new scaled 'screen bytes'
There are 256 possible screen bytes at 1x
2x needs a 512 byte Lookup Table
4x needs a 1024 byte Lookup Table

As MODE 0 only has 2 pixels per byte, we don't actually need the 4x lookup table in Mode 0

Building the Lookup Tables

Before we use the draw routines, we need to fill the LUT's...

The lookup tables are byte aligned

If the 2x LUT is at &16xx...
&16xx will be the First byte of the conversion for screen byte xx
&17xx will be the Second byte of the conversion for screen byte xx

The same is true for the 4x LUT... but it will use addresses &18xx &19xx &20xx and &21xx for the 4 output bytes

We need to mask each pixel of the source byte, and fill the destination byte with OR commands

We need to understand how the screen pixels work to do this... here we're building the 2x scale for Mode 1
It's pretty much the same for the MODE 1 4x scale routine.
The MODE 0 2x scaling routine is almost the same, just the mask and shifts are different.

There is no 4x LUT for Mode 0 - because there are only two pixels per byte, we can use the 2x LUT for the 4x version (we just use each entry twice)

Drawing the sprites!

We're going to draw our sprite!...

We have 3 different routines for each of the scales.

We use 'GetScreenPos' to calculate the screen ram location... this is the same as in the Simple Series.
Here's the 1x routine, we read each pixel from the source bitmap in HL... and transfer them to the screen at DE
When we want to scale the source sprite 2x we use the lookup table.

We read a byte as before from HL... but this time we load the lookup table into BC.

We swap C for the byte from HL - BC now points to the first conversion for that byte

We transfer the first byte from BC to the screen (This is the first of the 2x scaled equivalents of the byte from HL)

We increment B - BC now points to the second conversion for that byte

We transfer the first byte from BC to the screen (This is the first of the 2x scaled equivalents of the byte from HL)

We repeat the procedure for the next source byte.

The 4x procedure is the same... of course we use the 4x scaling LUT form Mode 1this time.

This time We transfer 4 bytes from the LUT to the screen

We INC B four times in MODE 2 for the 4 destination byte.

In Mode 0 we use the 2x LUT - and we write each byte from the LUT twice... INCing B twice - this is because there's only 2 pixels per byte, so there's no need for another LUT

We can also use the 4x LUT for 8X if we want super blocky sprites for some reason!


Lesson P62 - AMX Mouse reading on the Amstrad CPC

The AMX mouse is an 8 bit mouse supporting both the Amstrad CPC and the ZX Spectrum - it's a little unusual, rather than connecting via a dedicated mouse port, it in fact 'emulates' a joystick!...
When the mouse is moved up, the 'UP' button is pressed... the same for Down, Left and Right. and 'Fire' buttons
We could just play our regular game with the mouse, but really we need to read it during an interrupt handler for smooth movement... lets learn how!

CPC_Mouse.asm

Our Test program.

We're going to create a test program.

It will read the mouse via the joystick port and move a cursor around the screen.

When you click the left mouse button, it will draw a 'Chibiko' (our vampire mascot)

When you drag the left mouse button it will draw a blob trail.

For speed and simplicity, All sprites will be drawn with 'XOR'  meaning a sprite will be removed if it's drawn in the same position twice.

Reading the Mouse (Joystick)

We read the joystick in the same way as the keyboard - we need to select the row of the keyboard we want to read, and read that line in via the AY

We read in line &49 - the line for Joystick 1 - this contains all the joystick buttons the mouse use.

Keyboard Matrix

The Keyboard matrix uses 7 bits of line &49 for the Joystick - 4 directions and up to 3 firebuttons.

Fire 3 is undocumented, but is supported on the classic CPC as middle mouse - however it is NOT supported on the CPC+


      0            1            2            3            4            5            6            7     
&40 C-U C-R C-D F9 F6 F3 F-ENT F.
&41 C-L CPY F7 F8 F5 F1 F2 F0
&42 c [ RET ] 4 SHIFT \ c
&43 ^ - @ P ; : / .
&44 0 9 O I L K M ,
&45 8 7 U Y H J N
&46  6 J2-U   5 J2-D   R J2-L   T J2-R   G J2-F1   F J2-F2   B J2-F3  V
&47 4 3 E W S D C X
&48 1 2 ESC Q TAB A CAPS Z
&49 J1-U J1-D J1-L J1-R J1-F1 J1-F2 J1-F3 DEL

Our Mouse Routine

We're going to use a custom interrupt handler, this ensures mouse movements wont be missed if our program gets slow.
We're going to need some variables

We'll store the mouse position - we keep the last position so we know where the mouse was last time we drew it (so we can remove the old sprite)

We use 16 bit X,Y co-ordinates so we can have 'half movements' - we use the top byte of the X,Y pos for drawing...

Our drawing routine works in screen bytes - so the Mode 1 screen is 80x200 - meaning we need to move 1 unit in the X axis for every 4 or so in the Y axis.
We need to process each of the 4 directions... Up, Down, Left and Right

We check for each direction, and if it's pressed - we check the current location

If we're already at the edge of the screen, we don't want to move again, but if our current position is OK, we add or subtract the move as required and update the cursor position.
We also test the fire button, and if it's down we set our fire markers...

if it's released we clear the held marker. (The fire down marker should be cleared by the routine that uses the mouse)

This example just processes the left mouse button - we could mimic this for the right and even center buttons.

Just remember - Fire 3 (Center) will not work on the CPC+!

Using the Mouse Routine
We're going to use our mouse... First we test if fire was just pressed... if it was we'll show 'TestSprite2' (Chibiko) at our current location

We need to clear the Fire down event to mark it processed
We're going to update the cursor - first we check if the old position and new position are the same - if so we have nothing to do.

If the cursor has moved, first we redraw the cursor in the old position - as we're using XOR sprites this removes the old cursor,

Next we draw the cursor in the new position

Finally we copy the new cursor position into the old cursor position - for the next run of the main loop
Finally we'll process Fire held... this routine only runs if the cursor has moved.

If fire is held, we show a blob at the current position
We do NOT clear the fire held flag - it's cleared by the interrupt handler.


Lesson P63 - Kempston Mouse reading on the ZX Spectrum + SpecNEXT

The Spectrum can also use a mouse... the Kempston mouse - this is supported by the classic spectrum and spectrum next... lets put it to good use!

ZXS_KempsonMouse.asm
ZXN_KempsonMouse.asm



Mouse Routine
Reading the Kempston mouse is super easy!
Ww just read 3 ports

Reading &FBDF We can get an 'X position'
Reading &FFDF We can get an 'Y position'

Reading &FADF gets the Left and right mouse buttons in the format %000000LR
Here is some sample code to read the XY pos of the mouse in HL - and the Buttons in A


Note X and Y position read from &FBDF and &FFDF do not give positions on the screen - they are relative movements positions.

X=0 does not mean far left, nor does 255 mean far right... all we know is if the X pos goes down the mouse moved left... and if the Y pos goes down the mouse moved down.

Our Mouse Routine

We're going to need some variables

We'll store the mouse position - we keep the last position so we know where the mouse was last time we drew it (so we can remove the old sprite)
We use 16 bit X,Y co-ordinates so we can have 'half movements' - we use the top byte of the X,Y pos for drawing...

We also need two bytes to compare the last XY position to the current one (Xmove and Ymove)
First we test the fire button with port &FADF, and if it's down we set our fire markers...

if it's released we clear the held marker. (The fire down marker should be cleared by the routine that uses the mouse)
We read in the current X and Y position from &FBDF and &FFDF

We subtract these from the last read in value to get a movement distance - we flip the Y axis to match our screen co-ordinates.
We use these Xmove and Ymove values (now in Reg D and E) to alter the mouse position.

When we add or subtract the movement from each axis, we need to check if this causes an overflow,

If we go below zero, then we've gone over the top or left edge of the screen...
If we go over X>256-7 or Y>192-7 then we've hit the right or bottom of the screen.

We've now converted the movement to a valid onscreen mouse position

Using the Mouse Routine
We're going to use our mouse... First we test if fire was just pressed... if it was we'll show 'TestSprite2' (Chibiko) at our current location

We need to clear the Fire down event to mark it processed.
We're going to update the cursor - first we check if the old position and new position are the same - if so we have nothing to do.

If the cursor has moved, first we redraw the cursor in the old position - as we're using XOR sprites this removes the old cursor,

Next we draw the cursor in the new position

Finally we copy the new cursor position into the old cursor position - for the next run of the main loop
Finally we'll process Fire held... this routine only runs if the cursor has moved.

If fire is held, we show a blob at the current position
We do NOT clear the fire held flag - it's cleared by the mouse reading routine.



Lesson P64 - Mouse reading on the MSX
The MSX aslo has a mouse - it connects to the Joystick port, but is not read in the same way as a normal joystick.

We have to set some bits of the port, and we get a pair of nibbles for each X and Y axis - these are an 'offset' which we can use to build a 'position'... lets learn how.

MSX_MouseText.asm
MSX_Mouse.asm



Mouse Routine
Like the Joystick The mouse is read through AY ports A and B...

Unlike the Joystick, we need to send data on pin 8 to select the High/Low nibble of the axis.... our first read will get the X axis offset, the second will get the Y axis.

The bits of the ports are shown below

Port A (&A2 - AY Reg 14)
Bit Direction Details Joy Function Mouse Function
0 IN J1-Pin 1 / J2-Pin 1 Up Low/High Bit 0
1 IN J1-Pin 2 / J2-Pin 2 Down Low/High Bit 1
2 IN J1-Pin 3 / J2-Pin 3 Left Low/High Bit 2
3 IN J1-Pin 4 / J2-Pin 4 Right Low/High Bit 3
4 IN J1-Pin 6 / J2-Pin 6 Fire 1 Right Button
5 IN J1-Pin 7 / J2-Pin 7 Fire 2 Left Button
6 IN Key Layout Select (JP version only)

7 IN Cassette Read


Port B (&A1 - AY Reg 15)
Bit Direction Details Joy Function Mouse Function
0 OUT J1-Pin 6 Set to 1 Set to 1
1 OUT J1-Pin 7 Set to 1 Set to 1
2 OUT J2-Pin 6 Set to 1 Set to 1
3 OUT J2-Pin 7 Set to 1 Set to 1
4 OUT J1-Pin 8
Get Low/High Nibble
5 OUT J2-Pin 8
Get Low/High Nibble
6 OUT Port A Input Select Joy 1/2 Mouse 1/2
7 OUT Kana Lamp On/Off


Reading from the mouse

When we want to read a nibble from an axis, we need to send the configuration to port B to set up the mouse,

We then read from port A to get the data from the mouse
We need to send the correct data to port B - the bits we need to set are different depending if our mouse is in joystick port 1 or 2
The settings here are reading a mouse in joystick port 2 - use the REMmed out versions for joystick port 1

Here we're reading and combining the high and low nibbles - this gets one axis.
We run the GetMouseAxis command twice - this reads both the X and Y axis...

A will also contain the Left/Right mouse buttons in bits 4,5 - we set all the other bits to 1
Here's a simple routine to read in the mouse data, and convert it to an XY position in BC... A contains the LR Mouse buttons.

Note X and Y position read from here do not give positions on the screen - they are relative movements positions compared to the last read - even when we add them to previous reads, they are still only a 'relative movement - we need to covert this to a screen co-ordinate.

Our Mouse Drawing Routine

We're going to write a little mouse drawing routine,

It will show a cursor that's limited to the screen area, we'll be able to click to draw a 'Chibiko', and click and drag to draw blobs.

This is written on the MSX2
We're going to need some variables

We'll store the mouse position - we keep the last position so we know where the mouse was last time we drew it (so we can remove the old sprite)
We use 16 bit X,Y co-ordinates so we can have 'half movements' - we use the top byte of the X,Y pos for drawing...
First we test the fire button - it's still in the Accumulator, and if it's down we set our fire markers...

if it's released we clear the held marker. (The fire down marker should be cleared by the routine that uses the mouse)
We transfer the read in XY pos into DE
NOTE: The mouse reading routine has been flipped... a NEG command has been added to GetMouseNibble


When we add or subtract the movement from each axis, we need to check if this causes an overflow,

If we go below zero, then we've gone over the top or left edge of the screen...
If we go over X>256-7 or Y>192-7 then we've hit the right or bottom of the screen.

We've now converted the movement to a valid onscreen mouse position

This code uses MSX2 VDP commands to do the graphics work... if you don't know about them, you can learn more Here


Using the Mouse Routine
We're going to use our mouse... First we test if fire was just pressed... if it was we'll show 'TestSprite2' (Chibiko) at our current location

We need to clear the Fire down event to mark it processed.
We're going to update the cursor - first we check if the old position and new position are the same - if so we have nothing to do.

If the cursor has moved, first we redraw the cursor in the old position - as we're using XOR sprites this removes the old cursor,

Next we draw the cursor in the new position

Finally we copy the new cursor position into the old cursor position - for the next run of the main loop
Finally we'll process Fire held... this routine only runs if the cursor has moved.

If fire is held, we show a blob at the current position
We do NOT clear the fire held flag - it's cleared by the mouse reading routine.


Lesson P65 - Mouse reading on the Sam Coupe
Lets take a look at the Sam Coupe mouse.

It can be read from the keyboard matrix, but returns 12 bit values, lets learn how it works!

SAM_MouseCursor.asm
SAM_Mouse.asm


Generally, the tech manual"sam-coupe_tech-man_v3-0.pdf" is pretty awesome, but the documentation on the mouse is a massive pile of suck!... Either that or the author of these tutorials is too stupid to understand them (Either is possible!)

As the documentation was lacking, today's example was written by trial and error, it seems to work fine, but no promises can be made!

Mouse Port on the SAM
The SAM Coupe mouse is read from port &FFFE,

We need to do sequential reads from that port, and look at the bottom 4 bits.

We need to combine combine 3 nibbles to build the 12 bit X and Y movement.
Successive reads from &FFFE will return the following data:
Port &FFFE Bit
Read Number  7 
 6 
 5 
 4 
3 2 1 0
1 X X X X - - - -
2 X X X X - - - -
3 X X X X - R C L
4 X X X X Y-High Nibble
5 X X X X Y-Mid Nibble
6 X X X X Y-Low Nibble
7 X X X X X-High Nibble
8 X X X X X-Mid Nibble
9 X X X X X-Low Nibble
We need to read in 3 times from &FFFE to start, the last gets us the mouse buttons,

Now we do 3 reads, these will be the Y axis

Then we do 3 more reads, these will be the X axis

Our Mouse Routine

We're going to write a mouse reader which will draw a cursor onscreen.

Clicking the left button will draw a 'Chibiko' mascot, dragging will draw a trail of blobs,
The sprites are XOR, so drawing twice will remove a sprite.
We're going to need some variables

We'll store the mouse position - we keep the last position so we know where the mouse was last time we drew it (so we can remove the old sprite)
We use 16 bit X,Y co-ordinates so we can have 'half movements' - we use the top byte of the X,Y pos for drawing...
First read in 3 times and test the fire button with port &FFFE, and if it's down we set our fire markers...

if it's released we clear the held marker. (The fire down marker should be cleared by the routine that uses the mouse)

We use the DoAxis function to read in the Y and X move
We use these Xmove and Ymove values (now in Reg D and E) to alter the mouse position.

When we add or subtract the movement from each axis, we need to check if this causes an overflow,

If we go below zero, then we've gone over the top or left edge of the screen...
If we go over X>128-7 or Y>192-7 then we've hit the right or bottom of the screen.

We've now converted the movement to a valid onscreen mouse position

Using the Mouse Routine
We're going to use our mouse... First we test if fire was just pressed... if it was we'll show 'TestSprite2' (Chibiko) at our current location

We need to clear the Fire down event to mark it processed.
We're going to update the cursor - first we check if the old position and new position are the same - if so we have nothing to do.

If the cursor has moved, first we redraw the cursor in the old position - as we're using XOR sprites this removes the old cursor,

Next we draw the cursor in the new position

Finally we copy the new cursor position into the old cursor position - for the next run of the main loop
Finally we'll process Fire held... this routine only runs if the cursor has moved.

If fire is held, we show a blob at the current position
We do NOT clear the fire held flag - it's cleared by the mouse reading routine.



Lesson P66 - Quadtree video on the CPC (10k special) - Part 1/2
When my youtube channel hit 10,000 subs, I made a little demo which streamed a pre-rendered video

Lets take a look at the video compressor I wrote!

CPC_QuadTrees10K.asm



What are Quadtrees!?
Quadtrees are a way of breaking up an image into large blocks, then splitting those blocks into 4 smaller blocks when required.

This is intended to combine the advantages of large block sizes (Small bytesize for simple images) and the advantage of small blocks (flexibility for varied images)

Suppose we want to compress this image ->
First we split the image up into large blocks (in our example's case 64x64)

Each of these blocks will be considered
Lets take a look at this block

This block is quite varied, so we probably want to split it into four 32x32 blocks (Yellow).

The TopRight block is completely empty so we'll solid fill it.
The BottomLeft block is quite complex - maybe we should store it at high resolution

The TopLeft and BottomRight blocks are also rather varied, so maybe these should be split into four 16x16 blocks (blue), and considered separately!
This repeated splitting into quads is known as a quadtree.


The compression techniques
Because we're working on a lowly 8 bit system we won't be using Jpeg compression

To 'compress' the blocks, we'll simply store them at different resolutions

Something detailed, like the castle, we'll store at high resolution.

Something vague, like the mountain, we'll store at lower resolution.
We'll use the following possible compression algorithms:

1. Store the block as as solid filled color (or transparent)
2. Store the block at 1x1 (full resolution)
3. Store the block at 1x2 (half vertical resolution - only works for 8x8 blocks)
4. Store the block at 2x2 (half resolution - only works for 8x8 blocks)
5. Store the block at 4x4 (quarter resolution - only works for 16x16 blocks)



You'll notice most of our 'compression algorithms' only work for 8x8 or 16x16 blocksize... we'll have to split a block until we get to one of those sizes.
This may seem wasteful, but it also has advantages:
1. We can hardcode our program to work at a fixed size to gain speed.
2. Small blocks are more likely to be repeated in other places, so we can 'reusue' the bitmap data for two small blocks.


Don't be surprised if the resulting images aren't so great... We're stuck between some SUCK and a HARD place with this project!
As we're compressing hundereds of frames, We need to have speed and small sizes, so we're going to have to make tough choices!

The 1x2 resolution is chosen because it's easy to do on the CPC, and gives nearly as good quality as 1x1...
2x2 and 4x4 are done via lookup tables... we've covered that before.

Making the video

The video is split into 3 sections with 450 frames in total, and uses palette switching for extra effect!

The video is stored on a 481k CPC+ cartridge, which allowed for near instant bank switching, so no loading, and no 128k memory requirement.

In theory the video could be run from 512k ram in the same way, but the memory layout would need reconfiguring.
The Scene was rendered with the open source 'Blender' 3D rendering program.
The video was exported as a series of PNG files.
A special function was added to Akusprite Editor to load a video sequence.
The video can be exported with Save Quadtree Video

The default quality setting can be set in File format... this is used to choose the compression for each block

The default can be overridden on a per frame basis with the Setting option



The Quadtree routine

When we want to process a frame, we use the ProcessQuadTrees routine. This is a recursive routine which is also used when a block is split into quads

BC is the topleft corner of the block.
DE is the width and height of the block (in 8 pixel blocks, so &0808 is 64x64)
IX is the width and height in blocks... so &0403 is 04*64 x 03*64 = 256x192
the DoQuadTree routine will decide on the correct algorithm for decoding the block,

This routine is executed for each block in the area.
When we want to split a block into 4, we have an alternate version, to save a little time on loops.
To process a block, we read in a byte from the command datastream.

The top nibble defines the block type, the purpose of the bottom nibble depends on the block type.

Solid filled block

When we want to solid fill a block we have two routines to handle this

The color to fill the block is defined by the bottom nibble of the block definition.


A 'general' one for any size, and a fast '8x8 pixel' version.

Color 15 is the transparent color, so if we're filling with that, then this is a block to be ignored.

Otherwise we calculate the screen VRAM destination.
We use PUSH commands and stack misuse to fast fill.

We also do 8 consecutive lines 'unwrapped' to increase speed.

Working in 8 line blocks allows us to optimize our 'nextline' routine, as 7 consecutive 'nextlines' within a block are simple, but the 8th is more complex.
the 8x8 routine is even simpler, as we can just do 1 push (2 bytes - 8 pixels) per line, and not worry about getnextline


Full Resolution blocks

Bitmap blocks use a pixel data stream (PixelBlock), however so that repeated blocks are only stored once this is stored separately to the command stream.

The address in the bitmap stream is stored either as 11 bit, or 19 bit

This means there is either one address byte, plus 3 bits from the low nibble of the data block definition, or two bytes, plus 3 bits.

The bitmap data is aligned so The bottom two bits of the address are always zero, so a 2 byte address supports a range of 512k (for bank switching support)
The Fullres routine only works on 8x8 blocks, which makes screen address adjustment simple and removes the need for loops

We use stack misuse to quickly read data in from the source bitmap.


Half and Quarter Res

The Half and Quarter resolution compressors use a lookup table to quickly convert a byte in the stream to 2 or 4 screen bytes.

This was covered previously for sprites

Here is the Half res version

2 bytes are read in from the lookup table for each source byte.
A source byte is loaded from the stream, and this is used as the low byte  in the lookup table, the High byte is incremented to get the second destination byte.


Here we've looked at the concept of Quadtrees and how the blocks are decoded and displayed.

Next time we'll take a look at the CPC+ Cartridge, and some of the tricks we've had to do to get things working!

Lesson P67 - Quadtree video on the CPC (10k special) - Part 2/2
When my youtube channel hit 10,000 subs, I made a little demo which streamed a pre-rendered video

Lets take a look at the CPC+ code and cartridge routines.

CPC_QuadTrees10K.asm

Building our cartridge

We first compile our sourcecode to a binary file and then convert this to a CPR using BuildCPR.

Were including a modified OS,Basic and Amsdos rom which simulate a disk system using the cartridge. - Though our program doesn't technically need disk or basic support!

We're using BUILDCPR to covert these to a CPR
OS.ROM BASIC.ROM and AMSDOS.ROM were taken from the NoCart program, however BuildCPR gives us more flexibility.

However we have modified the ROM source of patchdos.asm with a new initialization routine.

Our 'bootstrap' is at memory address &dc00, however it transfers itself to address &A100 so that the rom can be paged out.

It pages in 3 banks of the cartridge, and copies &A000 bytes of the ROM to &0100 in ram, then executes address &0100.

Our program needs to run from this address and take control!
Our program starts at address &100

Thanks to the bootstrap it will be running in ram!

The Framelist

AkuSprite Exports the bitmap data for the frames and the Command Stream, However it also exports a 'template' of the image list with pointers to the addresses of the frame data.

This is intended to allow frames to be held onscreen longer, or repeated to save memory, however neither function was actually used.

BUT... We did inject some 'CPC+ palette switch commands' into the stream!



The Frame processing sequence goes through each frame in the Sequence.

First it checks if the frame is a palette switch (Speed=255) if it is, the address of the palette is loaded in and switched via the CPC+ functions.

Next it's checked if both bytes of the address are &FF &FF - this marks an end of stream (a real frame would never start at the end of a block!)


Anything else is a legit frame, so the bank is paged in and the frame is shown.


Once a frame has been shown we test the delay, and see if drawing the frame has taken longer than the delay, if not we wait a while until enough time has passed - this is to control the speed, as some frames are very slow, but others are super fast.


Rom Paging

There are 3 parts to our video.

1. The Framelist - This is used for the video sequence... it's always in memory.
2. The Command Stream - This defines the blocks that make up each data... It's paged in as needed
3. The Pixel Block data - This is used for bitmap blocks, it's separate from the commands so multiple blocks can share the same bitmap data... It's paged in as needed

We need to ensure 2 and 3 start in a new 16k bank... BUT we also need to offset these by 256 bytes, to compensate for the fact our program starts from address &0100, but appears in rom at the start of a bank &0000
The Frame routine includes a 1 byte 'bank number' so the correct bank can be paged in to process the Command Stream



If a frame spans over two banks, it can request a bankswitch to move to the next bank


The Two byte addresses for Pixel blocks include a 5 bit bank number (3 low bits in the low nibble + top 2 bits in the two byte address

CPC+ Palette
The QuadTree video decoder doesn't really use any CPC+ features (except rom switching) - so can be used on a classic CPC

However we use the CPC+'s more flexible palette to allow for nicer colors.

ChibiTracks for chibi beats!

The music of the demo is basically the same theme as the original ChibiAkumas game, however it has been converted to my new ChibiTracks music player!

ChibiTracks is a multiplatform player, which uses the ChibiSoundPro platform specific driver.

We play music during the interrupt handler.
ChibiTracks and it's requirements are included as ASM.
The interrupts on the CPC occur at 300hz... but our music needs to play at 50hz, so we skip 5/6 interrupts.

Chibitracks only uses AF,BC,DE,HL and IX - it does not use IY or shadow registers!
ChibiSound PRO and ChibiTracks support all the Z80,6502 and 68000 systems covered in these tutorials.

But We'll take a look at it in future lessons!

Lesson P68 - RLE+Bitplanes for lossless compression on the CPC
During my experiments with video, I also made a test of image compression. RLE compression works best with long strings of repeated bytes, and to try to increase their occurrence I split the image into 2 bitplanes. It's slow to draw, but saved a lot of memory!

CPC_BitplaneImage_Overscan.asm

Our bitplane data

We're going to compress an 'ovescan image' of 380x272.

The raw image is 27,278 bytes... if we RLE compress this it would go down to 13,425

BUT... if we split it into two bitplanes and RLE compress, the image will be a total of 11,144... saving 2,281 bytes for the identical resulting picture!
We're going to split the image into 2 bitplanes, and also as horizontal bytes are very likely to be different, but vertical ones may be more similar, we'll work in 8x8 blocks.

Coincidentally this is the same file format as the Gameboy uses!
The file needs to be RLE compressed, and split into two.

Akusprite Editor has an option to do this!
We'll include the two files in our source.

Our RLE compressor

We're using a flexible decompressor.

It uses IY as a pointer to the source file.
IXL is the bytecount for RLE data (all bytes same)
IXH is the bytecount for Linear data (all bytes different)

BC,DE,HL and shadow regs are unused.

call RLEDecompress to Start processing stream IY
call RLEGetNextByte to Move to the next byte - read from (IY) to get the result

Setting up our screen

We're going to set up the screen, we need to set up all the CRTC registers - you can see the full list here

We use port &BCxx to select register xx
We use port &BDxx to set the selected register to value xx

&0C/D define the base screen address, and also allow the 32k screen we need for overscan.
Here are the values we're using! Our screen is huge, and spans two 16k banks...
Bank 1 is &8200-&BFFF
Bank 2 is &C000-&FFFF
Memory <&8200 is free, so we'll use it for our stack!
Memory &0000-&3FFF is free so we can use IM1,
Memory &4000-&7FFF is free so we can use ram bank switching
but there's a slight memory 'gap' between them we'll have to deal with!
We're going direct to the hardware to set our colors.

We use port &7Fxx to do this.

We need to use hardware color numbers



Drawing our image
We XOR in our image data, so we clear the screen first
We point IY to our source data, and execute RLEDecompress to initialize the RLE data stream.

HL points to the VRAM destination
We work in 8x8 tiles.

First we go across the screen - 48 tiles total (384 pixels)

Then we go down the screen - 34 strips total (272 lines)
We'll do bitplane 0 - The top 4 bits of a byte are bitplane 0 in mode 1

We read a byte from the RLE stream (IY)

We take the top 4 pixels, mask bitplane 0 and write them to the screen... we move 1 byte to the right
We take the bottom 4 pixels, mask bitplane 0 and write them to the screen... we move back left

we then move down 1 line (by adding &0800 to HL) and repeat.

RLEGetNextByte moves IY to the next byte
We move across the screen 1 tile (2 bytes) and repeat until we've done an entire horozontal strip of tiles.
We need to move down 1 strip, and when we get to the end of the &8000-&BFFF bank, we manually step into the &C000 bank
We repeat until bitplane 0 is entirely drawn
We now do the same for bitplane 1 - The bottom 4 bits of a byte are bitplane 0 in mode 1

This example was just an experiment in saving a little memory by re-organizing data.

Take a look at the 10k subs example for more impressive 'lossy' compression - of course the video in that example is basically a sequence of images.

Lesson P69 - ChibiSound PRO on the CPC, MSX and Spectrum (128 - AY)
Lets take another look at sound!... we'll make an improved multiplaform sound driver for music!

CPC_V1_ChibiSoundPro.asm


ChibiSound PRO!

ChibiSound is the sound driver that handles the particularities of a system, there is typically one driver per system, though the CPC and MSX drivers are essentially identical except for the AY register setting routines.

The original 'ChibiSound' gave us one channel, one Volume bit, six pitch bits, and the ability to turn noise on. Pitches were not matched across systems, so sound 32 won't sound the same on all systems.

The updated 'ChibiSound Pro' gives us all the channels provided by the hardware, 8 volume bits, 16 pitch bits, and the ability to turn noise on. Pitches were not matched across systems, however the 'ChibiOctave' lookup table provides values which ARE matched across all systems.

ChibiSound PRO is essentially a reduced subset of AY functionality, and was designed on the Z80 - it's 'PRO' suffix is a parody of the 'SoundBlaster PRO' - which could only do 8 bit sound so wasn't up to professional standards! (neither is ChibiSound PRO)

ChibiSound PRO provides a standard interface to the underlying hardware, it allows the following features to be set for each channel on the underlying hardware:

Function
Register
Notes:
Channel Number (bit 0-6)
Noise On/Off (bit 7)
L Multiple channels can be supported, but on single channel systems only Channel 0 will be sure to play.
If possible Channel 0 will be a center channel, Channels 1+ may be left/right

Noise bit turns the noise effect on (1) or off (0) - this can be set on any channel, if the underlying hardware only supports one noise channel, this will be resolved by the driver.
Volume H Set volume of the channel (0-255). Higher numbers are louder. O is off
Pitch DE Set the pitch of the channel (0-65535). Higher numbers are higher pitch.
Using DE does not standardize the resulting pitch - however a 'Lookup table' of notes 'ChibiOctave' provides a standardized way of getting the correct DE value to get a pitch correct note on the platform.
Chibisound PRO does not offer features like Envelope, LFE etc, as providing consistent functionality across different platforms would not be realistic.


The new driver is a big improvement on the old one but doesn't really deserve the PRO suffix!

It's a parody of the early 'Soundblaster Pro' sound cards, which could only do 8 bit digital sound, so weren't really of 'pro spec' either!

AY Sound Chip

The AY uses a series of 8 bit registers to control the 3 sound channels.

To set a register, we must first select the register number by writing a byte to $FF8800, then send the byte of data to $FF8802

Register Meaning Bit Meaning Details
0 Tone Pitch L - Channel A LLLLLLLL Lower value = Higher pitch
1 Tone Pitch H - Channel A ----HHHH Lower value = Higher pitch
2 Tone Pitch L - Channel B LLLLLLLL Lower value = Higher pitch
3 Tone Pitch H - Channel B ----HHHH Lower value = Higher pitch
4 Tone Pitch L - Channel C LLLLLLLL Lower value = Higher pitch
5 Tone Pitch H - Channel C ----HHHH Lower value = Higher pitch
6 Noise Generator ---NNNNN Higer = Faster noise
7 Mixer  --NNNTTT   N=Noise T=Tone (Channel --CBACBA 1=mute 0=normal)
8 Amplitude - Channel A ---EVVVV E=Envelope (1=Enabled) VVVV=Volume
9 Amplitude - Channel B ---EVVVV E=Envelope (1=Enabled) VVVV=Volume
10 Amplitude - Channel C ---EVVVV E=Envelope (1=Enabled) VVVV=Volume
11 Envelope L (Volume over time)  LLLLLLLL Lower=Faster Envelope
12 Envelope H (Volume over time)  HHHHHHHH Lower=Faster Envelope
13 Envelope Selection ----EEEE Envelope number (See PDF)

For more details, please see the AY sound chip PDF

Only the way we send data to the AY chip changes on each of the 3 platforms, the main sound code is identical!

The ChibiSoundPRO source - All platforms

We need some parameters!

We have some cache ram to keep track of the AY register settings
We have a 'ChannelMap' which allows us to simulate 8 possible virtual channels

We have a 'ChannelMask' to get the bit pattern for the channels Mixer settings
DoChannelMask Will get the settings to set the current channels mixers bits.

it also points Point BC to previous mixer value in ram, in case we need it!
ChibiSoundPro_Init will set things up, loading the default 'silent' value into the mixer cache
ChibiSoundPro_Set will configure our sound channel

H=Volume (0-255)
L=Channel Num (0-127 unused channels will wrap around) / Top Bit=Noise
DE=Pitch (0-65535)

The channel number can be up to 128, so we remap it to channels 0-2 with the ChannelMap table
Now we check the volume - if it's zero we want to silence the channel.

We do this by setting the channels Noise and Tone bits to 1 in the mixer (reg 7)
We need to send the 12 bit frequency to the two registers for the channel

Channel 0 uses 0/1
Channel 1 uses 2/3
Channel 2 uses 4/5

We also need to flip the bits, as chibiSoundPro uses the value &FFFF as the highest frequency, but the AY uses &-000
If we need to make a noise, we need to clear the channel tone bit in the mixer, and clear the noise bit to turn noise on

We also set the 5 bit  noise frequency in reg 6
If we want to make a tone...

 we need to set the noise bit (turning noise off)

we need to clear the tone bit (turning tone on)
Finally we set the volume.

The AY uses a 4 bit volume, and this is set for the 3 channels with Reg 8/9/10

Platform Specific AYRegWritePro

AYRegWritePro will set AY register A to value C

it uses AYCache to avoid writing values which are unchanged to the hardware.
On the CPC we have a rather odd sequence of commands! this is because the AY chip is linked through the keyboard hardware.
on the Spectrum we use ports &FFFD and &BFFD to select the register and send the new value
on the MSX we use ports &A0 and &A1




Back to Page 6