Notes on the instruction set

Note

This section applies mainly to the Digirule2/Digirule2A instruction set. The main ideas about providing an efficient instruction set for a CPU that does not have a lot of RAM available still apply to the Digirule 2U as well of course.

This section contains a few notes on modifying the instruction set of a Digirule2, so that it becomes more efficient. Since the Digirule2 has limited memory, it does pay off to implement slightly more complex instructions to save memory and this is what is meant here by “efficiency”. Saving bytes of memory while achieving the same end result.

For example, the Digirule2 instruction set contains both a SUBLA and an ADDLA instructions when it could only have offered an ADDLA and let the user perform subtractions in two’s complement by inverting the second operand.

Here is what this looks like when trying to calculate 4-2:

With the availability of SUBLA:

COPYLA 4
SUBLA 2
HALT

Listing A

Without the availability of SUBLA:

COPYLA 2
XORLA 0xFF
ADDLA 4
HALT

Listing B

Listing A is 5 bytes but listing B is 7 bytes long. In this case, having a separate instruction for addition and subtraction helps to fit more instructions into the already limited memory space of the hardware.

This however is not always the case and the following sections offer some suggestions for improvement.

Throughout those:

• The * character denotes “any character” when it is not used to denote multiplication. For example, SHIFTR* is meant to include both SHIFTRL, SHIFTRR instructions.

• addr denotes an address within the address space of the Digirule2

• mem[addr] denotes the content of addr within the address space of Digirule2.

Memory Operations

Indirect Copy

As demonstrated in section Advanced Digirule 2 Programming, there is a clear need for an indirect copy instruction. That is, a copy instruction that can copy between memory offsets stored in memory.

Currently, COPYRR addr1 addr2 is a 3 byte instruction that performs mem[addr2] = mem[addr1]. Unless COPYRR is written as a subroutine, it is impossible to get it to copy between two addr1, addr2 that are the result of a calculation. However, doing so requires the following pattern:

COPY** f_crom
COPY** f_to
CALL f_copy

f_copy:
.DB 7
f_from:
.DB 0
f_to:
.DB 0
RETURN

The first two copies denote that f_from, f_to will definitely have to be populated prior to performing the copy and are therefore required, irrespectively of which form they take. They usually are COPYRR variants though, which means that these two copies cost 6 bytes already, plus 3 bytes for the main COPYRR (denoted here by f_copy), plus 3 bytes for the costs of CALL and RETURN, not to mention the cost of calling the function.

That is, 12 bytes of memory to implement an indirect COPYRR. The alternative here would be to have a COPYRR variant that instead of implementing mem[addr2] = mem[addr1], it implements mem[mem[addr1]] = mem[mem[addr2]]. In that case, the memory cost would be just 3 bytes.

Arithmetic Instructions

Target of SHIFT**, BCRS*

Most of the mathematical operations of Digirule 2’s instruction set target the Accumulator. For example, addition, subtraction (ADD*A, SUB*A) and the elementary logic operations or AND*A, OR*A, XOR*A require one of the operands to already be in the accumulator.

The only exception to this are the shifting and bit testing operators.

This means that if a calculation involves an intermediate step where the value of an operand has to be shifted, the current value of the accumulator has to be copied to a memory location, shifted there and copied back to the accumulator to continue with the rest of the calculation.

Both copies would be performed via a COPYAR, COPYRA which means a potential loss of 4 bytes of memory, if the calculation cannot be expressed in a different way.

The suggestion here is to have variants of bit testing and shifting that can target the Accumulator too.

SHIFTRR, SHIFTRL

These two instructions shift bytes left or right and are equivalant to division or multiplication by 2, respectively. On the Digirule 2, shifting is performed through the Carry flag. If a program is performing a series of operations and it only calls for a plain right or left shift, the Carry flag has to be manually cleared so that it does not interfere with the result of the calculation. This inserts 3 bytes for each CBR instruction that ensures that the Carry flag is clear prior to shifting.

One practical example is provided in the Pseudorandom Number Generator (PRNG) that uses a plain Linear Feedback Shift Register. In this technique, it is required to shift and XOR the current state of the PRNG to calculate the value of the bit at its input.

Therefore, in cases like these, where only a shift is required, offering a plain SH* instruction would help in conserving memory.

Flow control instructions

Indirect JUMP and CALL

Similarly to the reasoning of the indirect version of COPYRR, an indirect version of JUMP addr, CALL addr would simply jump to memory location mem[mem[addr]].

To an extent, this is already implemented currently through ADDRPC but requires an addition and it also does not hint at a JUMP operation.

Being able to transfer execution in such a way would also enable functions to be passed as parameters to other functions.

Therefore, the suggestion here is to add indirect versions of these two instructions.