Isetta TTL computer

Project Logs

Collapse

Another interesting BASIC to test the 8080 instructions
roelh • 05/23/2023 at 18:52 • 1 comment
Since the Isetta supports instructions of the Z80 processor, it also supports the famous 8080 processor, because the Z80 has the 8080 instruction set as a basis. The 8080 is interesting because it is the grandfather of almost all INTEL CPU's for personal computers of the past 40 years.

One of the very first personal computers was the Altair 8800 from the company MITS. It was built around the 8080 processor. Bill Gates, Paul Allen and Monte Davidoff created a BASIC interpreter for this computer, the Altair BASIC. It was called a 4KByte BASIC. But the actual program was smaller than 4K, because 4K was all memory that the Altair had ! There were only 790 bytes free for user programs ! Altair BASIC was the very first product of Microsoft !

So I used this Altair BASIC to test the 8080 instructions on the Javascript emulator of the Isetta processor. I found an Annotated disassembly and a binary version. And I really needed the disassembly, to understand a little of the program.
Subtract and compare

After the first tests, I found that for the 8080/Z80, the carry flag behaviour for the subtract and compare instructions is inverted w.r.t. the behaviour in the 6502. I created a new microinstruction that complements the carry flag for the 8080/Z80. In several situations, it can be executed at the same time as another microinstruction.
Parity flag
Altair Basic has a FCompare at 0A4C that compares two floating point numbers. It returns
one of the following values:
- 0x00 for zero
- 0x01 for greater than zero
- 0xFF for less than zero
That's very logical. Now you can simply test the sign flag or zero flag to jump according to the result.
But in two instances, at 0B63 and 0C03, it is followed by a JP PO (jump on parity odd) instruction,
that will jump if the result is greater than zero (as 0x01 is the only one of these three values that has an odd number of '1' bits in the byte)
On the 8080, (almost ?) all arithmetic and logic instructions that act on the accumulator will
set the P flag according to the parity.
But on the Z80, the parity flag behaviour is different. Logic instructions will set the flag according to the parity, but the arithmetic instructions will set the flag when there is an overflow. And indeed, it has been reported on the Internet that the Altair Basic will not work on a replica Altair that has a Z80 processor instead of the 8080 in the original Altair.
So, although it is widely believed that the Z80 can run all programs intended for the 8080, this
is not true because the behaviour of the P/V flag is different.

And on the Isetta it is also a problem because parity is not implemented at all.
The parity flag is tested with JP PO on only two occasions, at 0B66 and 0C04. Both instances are preceded by a call to FCompare at 0A4C. I replaced the calls to FCompare by a new subroutine, that does the following:
- Call the FCompare function at 0AC4
- Return if result is non-zero
- Save Accumulator
- Load Accumulator with 0xFF, and OR A to set the sign flag
- Restore Accumulator and return
Both JP PO instructions are now replaced by JP P (jump if positive). The parity flag is not needed any more (at least not for this BASIC).
A nasty bug
Strange thing were going on. The display of numbers displayed the first digit as a letter. And with another test program (TRS-80 Basic), identifiers were not recognized. After a few days, I found out that the carry flag was not cleared after an XOR A,A instruction. The microcode seemed ok, it said the carry was written (and at the same time a load operation was done). But for the carry-complement (for subtract and compare) I changed the behaviour such that writing the carry while doing a load resulted in a complement-carry ! Sometimes it is difficult to program your own contraptions.
Other changes to the BASIC
The code at the following positions was changed for character output and keyboard input:
- The code at 0377 that handles character output
- The code...
Read more »
Javascript emulator runs Apple 1 BASIC
roelh • 05/09/2023 at 18:43 • 0 comments

The emulator is ready (in Javascript), and the microcode that makes Isetta a 6502 was also completed. The emulator mimics the operation of the hardware, so it needs the same microcode.

What could be better to test it, than the original BASIC of the Apple 1 ?

There were a few mistakes in the microcode, the BIT instruction did not set all flags correctly, and the LDX/LDY-immediate did not set the flags. And originally I thought I could do without the overflow flag, but the Apple BASIC uses it, so I changed my mind and also implemented this.

Here is the proof that it is working:

After the basic is loaded in the memory of the emulator, a few patches are done to enable keyboard-input and textarea-output. It is amazing that you must jump through several hoops to convert javascript-keystrokes in 'normal' ASCII.
First 6502 microcode
roelh • 04/22/2023 at 10:16 • 0 comments
A first version of the microcode was made. The microcode is generated with a Javascript program. The script can be downloaded from the file section. An effort was made to make the microcode as clear as possible, the output will be a list of microcode instructions with a good explanation, for all 151 opcodes of the original 6502. It lookes like this:
```
---- 0xC8 INY reg_y ----
001900 3A84B3 3A84B3 to_t <- inc(reg_y),upd_nz
001902 008301 008301 next
001904 11B403 11B403 reg_y <- acc_t

---- 0x69 ADC imm ----
000D20 D1AB80 D1AB80 to_dpl <- (pc++),irq_to_f
000D22 5D46F0 008301 F:[to_a <- adc(acc_a, dpl),upd_nzc] T:[next]
000D24 000401 5D46F0 F:[interrupt] T:[to_a <- adc(acc_a, dpl),upd_nzc]
000D26 110400 110400 int2

---- 0x65 ADC zp ----
000CA0 D1AB80 D1AB80 to_dpl <- (pc++),irq_to_f
000CA2 5DC5F0 008301 F:[to_a <- adc(acc_a, (0|dpl)),upd_nzc] T:[next]
000CA4 000401 5DC5F0 F:[interrupt] T:[to_a <- adc(acc_a, (0|dpl)),upd_nzc]
000CA6 110400 110400 int2

---- 0x75 ADC zpx ----
000EA0 D18B80 D18B80 to_t <- (pc++),irq_to_f
000EA2 1DA402 1DA402 to_dpl <- add(acc_t, reg_x)
---- end of ea calculation ---
000EA4 5DC5F0 008301 F:[to_a <- adc(acc_a, (0|dpl)),upd_nzc] T:[next]
000EA6 000401 5DC5F0 F:[interrupt] T:[to_a <- adc(acc_a, (0|dpl)),upd_nzc]
000EA8 110400 110400 int2
```
You can easily generate the full microcode:
- Go to: this W3Schools page.
- Clear the text in the left window and paste the microcode-generator code there.
- Press Run
- The microcode will appear within a few seconds.
Some remarks about the notation:
- The first hex number on a line is the address in the microcode ROM
- The next two hex numbers are the microcode. These numbers are the same, except when conditional execution is used
- When there is no interrupt, only the T:[ ... ] sections are executed.
- When there is an interrupt, the F:[ ... ] sections are executed.
- 'next' is an indication that the next opcode will be loaded. Due to the pipeline effect, one extra instruction will be executed after 'next'.
- Depending on its role in the instruction, register A will be shown as 'to_a' or 'acc_a'. Same for register T.
It is easy to count the number of cycles for each opcode. Just count the lines up to the first instruction that contains 'next', then add one cycle for the instruction after 'next'. So the shown 'ADC zp' opcode takes 3 cycles.

The operation of some instructions will probably still be unclear to you. In that case, check the microcode script that contains a lot of comments about the used instructions.
In the microcode generator, there is a central role for the function 'ins6( a, b, c, d, e, f )'. The name 'ins6' stands for 'instruction with 6 components'. The six components are:
- a. Condition. An instruction to move a condition to the F flag, like c_to_f, z_to_f, or irq_to_f.
- b. Destination. Data destination, like to_t, to_a, to_dpl.
- c. ALU operation, like ld, adc, and, rol.
- d. Source 1, this can only be acc_a or acc_t. For single-operand instructions, source 1 is unused. For writing to memory it decides if A or T gets written.
- e. Source 2, this will in most cases be a memory location, like 'dir' (zpage location), 'ext' (full 64k range address in dph|dpl), 'pc_pp' ( (pc++) ), or a hardware register (dph, dpl, pch, pcl), or a register location in memory (like reg_s, reg_x), or a small constant lit0, lit2, lit8, lit32 for the values 0, 2, 8, 32. For writing to memory, source 2 is the destination address.
- f. Update flags. It is indicated as upd_nz, upd_c, upd_nzc. Any combination of N, Z, C flags is allowed.
The ins6 simply combines the control-wire values of its arguments, so it would not matter in which order the components are placed. But the comment-printing function 'pr_print' depends on the order of this components.

It is tempting to continue with writing the Z80 micro-instructions, but I will now first make a simulator to check if everything works as intended.
Control section details
roelh • 04/15/2023 at 21:15 • 0 comments
In this log I talk about the 24 control signals that come from the microcode. The names of all signals that come from the microcode register will start with CTL.

The microcode will be organized in 3 bytes, the upper, middle, and lower control byte. The upper microcode has the longest description.

MICROCODE UPPER CONTROL BYTE

The first diagram shows the signals that select a condition and handles flags:

Flag update

Some instructions update flags, others don't. The signals CTL_UPD_C, CTL_UPD_Z and CTL_UPD_N indicate if the flags C, Z or N must be updated with a new value coming from the ALU_COUT, the ZERO detector at the output of the ALU, or bit 7 of the ALU output (R7, Result-bus bit7). If a flag is not updated, it will keep the same value.

There is no provision to read or write the flag bits as a full byte, like that is normally used in many 6502 designs. The microcode will have to read or write the flags one-by-one in order to push or pull the flags (or do EX AF,AF' for Z80).

Condition select

The signals CTL_C0, CTL_C1 and CTL_C2 select one of eight conditions. It is used for two different things:
- As a selector of the CONDITION carry-in signal of the ALU (adder or shift-right)
- As a selector for the FLAG_F for conditional execution. This second function is activated when CTL_WR_F is active (high). If CTL_WR_F is low, FLAG_F keeps the same value.
Early condition evaluation

There is one more important thing to say about the condition selection. Suppose that we want tot test for a condition, and execute a conditional micro-instruction based on that, this would be 3 micro-instructions (3 cycles):
1. Select the relevant condition, like C, Z or N, (putting this condition in FLAG_F)
2. Fetch the next micro-instruction based on the FLAG_F value
3. Execute the micro-instruction that was selected with FLAG_F
Number 1 and 3 are really needed, but in number 2 we are just waiting for the next micro-instruction coming out of the pipeline, and if we can not do something useful in that cycle, it is a 'lost cycle'. But we can do something about that. By connecting the signals CTL_C0, CTL_C1, CTL_C2 and CTL_WR_F not to the microcode register, but directly to the microcode ROM output, we can evaluate the condition one cycle earlier (at the end of the fetch phase). We now get the following two-cycle sequence:
1. Select the condition. The selection has actually been done in the previous cycle, so in this cycle the new micro-instruction, based on the FLAG_F value, will be fetched.
2. Execute the micro-instruction that was selected with FLAG_F
So we just gained one cycle for micro-instructions that use conditional execution.

ALU CONTROL

The ALU control section must provide the control signals for the ALU. In the ALU log it is shown that the following control signals are needed:
- 4 bits DCBA controls for the upper logic unit
- 4 bits DCBA controls for the lower logic unit
- 2 bits Carry-input signals for added (ADD_CIN) and right-shift (SHR_CIN)
- 1 bit selection of shift-right unit CTL_SHR
So the ALU can be controlled with 11 signals. But we don't want to spend so many of the available micro-instruction bits for these. Fortunately, it is not difficult to bring this number of signals down to a reasonable amount.

Generating the DCBA control signals

The DCBA controls for the upper logic unit have bit A being '1' only in the case of a DEC instruction (and for complementing a value, but we don't use that). So bit A can be removed from the control wires and be regarded as a special case.

The DCBA controls for the lower logic unit only use the patterns 0000, 1010 and 0101. Of these, the 0101 is only used in a single case (SUB, Subtract), so if we handle that as a special case we only need a single control signal CTL_ALU0 for choosing between 0000 and 1010.

To handle the special cases, we have a decoder (U5) that decodes the CTL_ALU1, CTL_ALU2 and CTL_ALU3 lines into 8 separate cases. The output of the decoder...
Read more »
CPU control section
roelh • 04/14/2023 at 20:45 • 0 comments

The following diagram shows the principle of the control section:

START OF NEW INSTRUCTION
As said before, the processor is controlled by microcode. At the start of a new 6502 or Z80 instruction, the opcode of that instruction will be put in the 8-bit instruction register and the counter will be reset to zero. The counter has 4 bits, so for each opcode in the instruction register there can be a sequence of maximal 16 micro-instructions.
SEQUENCE

The Microcode ROM will deliver the 24-bit micro-instruction and store it in the microcode register at the end of the cycle. At the end of the cycle, the counter is incremented to prepare for fetching the next micro-instruction. In the next cycle, the micro-instruction is decoded (in the decoding block) and the micro-instruction is executed.

For each opcode, there is a unique sequence of micro-instructions that will be executed. At the end of this sequence there is a special micro-instruction called LD_IR that loads the new opcode in the instruction register, resets the counter, and sets the 3-bit page.
PIPELINE

Note that when the micro-instruction is executed, the following micro-instruction is already being looked up in the microcode ROM. This means, that when the decoding section decides that a new opcode must be put in the instruction register, the next micro-instruction is already read from the microcode ROM. So the LD_IR is not the last instruction in the sequence: The micro-instruction that follows LD_IR will also be executed. This is the pipeline effect.
CONDITIONAL EXECUTION

The control unit is capable of conditional execution of micro-instructions. At each location in the microcode ROM, there are actually two instructions stored. Which of these is executed, is determined by a flag called F. If, for a certain instruction, we do not want it to be conditional, we simply store two identical instructions, so in that case it doesn't matter which one is executed.
PAGES

If we would only have 6502 instructions, the page register would not be needed because there are only 256 possible opcodes (of which many are unused). But a Z80 would need several pages, because in its basic set of 256 opcodes there are some that are followed by another opcode byte. It is expected that the Z80 can be handled in 4 pages (A basic page, index-IX-IY page, shift-and-bit page, and the 0xED page for the special Z80 instructions). So of the 8 available pages, 1 will be needed for the 6502 and 4 for the Z80. The page register can be changed by the LD_IR instruction.
Requirements for the control section
roelh • 04/06/2023 at 20:21 • 0 comments
Micro-instructions... which instructions do we need ? Before the control section can be designed, we need a list of everything that we want to control with a microinstruction.

We start with a short list of pseudo-code that will have to be supported, it tells how the information will flow and what the ALU does:
```
// Load Instruction Register with next opcode. It would be nice 
// if it can also be loaded from another source. This micro-instruction must
// also specify which microcode page to use for the next opcode. Each microcode 
// page has room for 256 opcodes of max. 16 microinstructions each.

    IR <- (PC++)

// Instructions with a single operand.
// op1 can be LOAD, INC, DEC, SHR

    dst_reg <- op1 (data_source)

// Instructions with two operands. Can use A or T register as first operand.
// op2 can be ADD, SUB, ADC, SBC, AND, OR, XOR

    dst_reg <- op2 (A, data_source)
    dst_reg <- op2 (T, data_source)

// Store instructions. Only A or T can be stored.

    mem(address) <- A
    mem(address) <- T
```
The dst_reg can be A, T, DPH, DPL, PCH. When PCH is written, DPH will be written to PCL

The data_source can be:
- a register (DPH, DPL, PCH, PCL),
- a small constant (CGL)
- a value read from a memory address.
A memory address can be:
- a register stored in memory, addressed by the CGL constant as low byte and a zero as high byte
- a zero page address (for 6502), addressed by the DPL register, with high byte zero
- a full 64k address, provided by DPH and DPL
- the program address, addressed by PCH and PCL. Automatic increment of PCH/PCL after it is used.
The register that is stored in memory sits in a special 64K memory area called SYS. It is addressed by the CGL constant (4 bits). The lower 3 bits specify 8 registers. If the 4th bit is set, it specifies 8 other registers, but in this case there is an extra address bit that is provided by the Z80-exx flipflop. The result is that the upper 8 registers can be swapped for other registers by toggling this flipflop. It implements the alternative register set of the Z80.
ALU FUNCTION
The control section has to tell the ALU what it should do:
Two-operand instructions like ADD, SUB, AND, OR, EOR
One-operand instructions like LD, INC, DEC, ASL, LSR, ROL, ROR

FLAGS

But the microinstruction has to specify more:
- which processor flags ( N, Z, C ) to update
- if the internal carry flag (TC, Temporary Carry) must be updated (see below)
- which value to use for the carry-input of the ALU and for the SHR block ( 0, 1, C, TC )
- which condition to use for conditional execution (see below)
The Temporary Carry is a flag that is used for address calculations, in situations where the programmer-visible C flag may not be changed.

CONDITIONAL EXECUTION

The control unit is capable of conditional execution of microinstructions. At each location in the ROM that stores the microinstruction, there are actually two instructions stored. Which of these is executed, is determined by a flag called F. If, for a certain instruction, we do not want it to be conditional, we simply store two identical instructions, so it doesn't matter which one is executed.

This flag F can be modified by the microcode. For example, in a BCS (branch on carry set) instruction, the flag F will first get the same value as the C flag. The following microinstructions will now put the jump-address in the program counter when the F flag is '1'.
Another use is, that the Interrupt-signal can be copied to the F flag at the beginning of the microcode for an instruction. At the end of the microcode sequence, the F flag will then be tested, and if there was an interrupt, the instruction register (IR) will be loaded with the opcode of the instruction that handles the interrupt. (That will probably be an opcode in an other page.)
The F flag can be set to one of the following conditions:...
Read more »
The ALU of Isetta
roelh • 04/01/2023 at 14:35 • 1 comment
Which functions do we need in the ALU and how must that be arranged? Well, we need the usual ones:

Arithmetic:
- ADD (Add two bytes)
- SUB (Subtract one byte from the other)
- INC (Increment: Add 1 to a byte)
- DEC (Decrement: Subtract 1 from a byte)
Logic:
- bitwise AND
- bitwise OR
- bitwise XOR (called EOR for the 6502)
Shifts:
- Shift the bits in the byte one bit to the left (For ROL and ASL on the 6502)
- Shift the bits in the byte one bit to the right (For ROR and LSR on the 6502)
Pass:
- PASS Pass data from databus unmodified.
It must be possible to pass a byte from the databus to the output of the ALU, because otherwise it would not be possible to simply load a value in one of the registers (it is needed for LOAD functions).

The ADDER

The ADD function is easy to realize. There is a chip that can add two 4-bit values, it is called the 74AC283. If we use two of them, we can add two bytes that are present on the A and B input of the adder. The chip has a 'carry' input that can be connected to the Carry flag to provide ADC (Add with carry).

We need a way to connect the A-input of the adder to zero, this causes the B-input (databus) to be added to zero, passing the databus value to the output, providing the PASS function. If we set the carry-input of the adder to logic 1, this function will become INC.

If we connect the A-input of the adder to the value 0xFF (all 8 bits '1'), the adder will add the value 255 to the B-input (databus). But this is a value greater than 8 bits ! The 8th bit will be dropped and the result is that the databus value is decremented, so we now have DEC.

And we need a subtract function, SUB. This can be calculated by bitwise complementing (replacing 0 by 1, and 1 by 0) the B-input, and then ADD (while providing '1' on the carry input). The 'carry' input can be connected to the Carry flag to provide SBC (Subtract with carry).

Left shift can be done by adding a value to itself. The microcode will handle that, so the 6502 or Z80 functions to shift left will work as usual. The right shift needs a dedicated chip.

A simple way to do LOGIC functions

For providing logic functions, the easiest way to do that seems to be to build something like the following circuit:

This not only provides the logic functions, but also ADD. It is very easy to understand, the left side of the diagram calculates the ADD, AND, OR and XOR functions, and at the right side the ALU Opcode selects which of the four results will be used (with a 4-input multiplexer). It will work perfectly.

How many parts will be needed ?

Since this is an 8-bit ALU, we need 8 gates of each type. There are 4 gates in a chip, so calculating the three logic functions cost 6 chips. We need eight 1-bit multiplexers, there are two in a chip, so that cost 4 chips, and there are the two adder chips, for a total of 12 for the circuit above.

But there are more chips needed. For subtract, the B input must be complemented, that can be done with two 74HC86 XOR gate chips. And for INC or DEC we must be able to put a value of 0 or 0xFF on the A input. This can be done with two multiplexers 74HC157, that can connect the A-input either to the input bus or to the fixed value 0 or 0xFF. That is 4 chips for these special functions, bringing the total to 16 chips for the ALU.

Calculating Logic functions with less chips

In the previous section we used a multiplexer. It has two inputs X and Y that select one of the inputs A, B, C or D and put the selected signal on its output Q:

The truth table of this device is:
```
X   Y   Q
----------
0   0   A
0   1   B
1   0   C
1   1   D
```
If we now regard X and Y as inputs and A, B, C and D as constants, we actually have a programmable logic gate, where the value on the four ABCD inputs determines which 1-bit function it performs on the values X and Y:
```
DCBA
0000  Always 0
1000  AND
1110  OR
0110  XOR
1100  pass value X
1010  pass value Y
0011  pass value X complemented
0101...
```
Read more »
Data flow in the CPU
roelh • 03/31/2023 at 20:28 • 0 comments

The following diagrams will clarify how the information flows in this CPU.

All registers and busses that are shown here are 8 bits wide.

INSTRUCTION FETCH
This diagram shows how the opcode of an instruction is fetched from memory. The program counter, consisting of an 8-bit high part (PCH) and an 8-bit low part (PCL), addresses a location in memory where the opcode can be found. This opcode is read from memory and transferred to the instruction register (IR). The instruction register is connected to the control unit, that is responsible for the execution of this instruction. It will do this by executing micro-instructions that are in its own private read-only memory (not shown here). The program counter (PCH/PCL) contains a counter that will increment after the opcode is read.

OPERAND FETCH
If the instruction has an operand, the operand will be addressed by PCH/PCL. It will be read from that address in memory and will first be transferred to the databus and the ALU. The ALU is in this case configured to ignore the contents of the operand bus, and transfer the contents of the databus to its output without modifying it. The byte then is transferred through the result bus and delivered to one of the indicated registers A, T, DPH, DPL or PCH.

ADD FROM ZERO PAGE TO A
This example shows how an operand in zero page is added to register A, with the result flowing into register T. It supposes that the opcode of the instruction has been read, and that an operand (the zpage address) has been read and has been put into register DPL.

We see here that the lower 8 bits of the address come from DPL while the upper 8 bits are zero. The contents of that memory location is read from memory and transferred to the input of the ALU. The other input of the ALU is equal to the contents of the A register. The ALU will be configured to ADD both inputs, and the result of the ALU is transferred to the T register.

INCREMENT DPL
This diagram shows how one of the internal registers (DPL in this case) can be incremented. The ALU is configured to transfer the databus to the result bus (ignoring the operand bus), but the ADDER inside the ALU has its carry-input set to 1, so the ALU will increment the value that is presented on the input.

STORE REGISTER A
Here is shown how the A register is stored in memory, at a memory location that is addressed by registers DPH and DPL.