The armlet instruction set

An instruction – a symbol in the alphabet of the machine

To be able to eventually program the armlet architecture, we need a language that we can use to instruct the processor what to do. For example, to configure the data path, one possibility is to specify values to all the control inputs of the data path individually, but this requires many bits of input, many of which are redundant. For example, if we are setting up an arithmetic operation with register inputs only, both the memory interface unit and the immediate data bus will be inactive. Similarly, the arithmetic logic unit will be inactive if we are setting up a memory load or store.

A common unit of measure in terms of the number of bits used to configure a circuit is the word length. The armlet design is simple enough so that 16 bits (the word length) easily suffice to give a unique binary encoding to each operation in the set of all possible operations in the design, except of course those operations that take an immediate data input, which in itself takes 16 bits of data. Such a binary word that configures the processor is called an instruction.

From a human perspective it is also convenient to have an equivalent symbolic representation to the binary representation. For example,

add $4, $0, $3

is more human-friendly than the binary word

0011000100000110,

even if it is actually the binary word that configures the processor. The set of all available operations and their binary representations is called the instruction set of the architecture.

Binary representation of the armlet instructions (**)

The armlet architecture has two types of instructions. The first type does not take an immediate data operand. All these instructions are encoded in binary as one word (16 bits). The second type includes an immediate data operand. All such instructions are encoded in binary as two consecutive words (16 + 16 = 32 bits), one word to set up the operation, and one word for the immediate data.

The possible operations that we have discussed until now in fact constitute the almost complete instruction set for the armlet architecture. The only instructions that are yet missing are the control and branch instructions, which we will review in Round 5: A programmable computer.

The armlet instructions are encoded in binary using a bit-packed format based on a 6-bit opcode (stored in the least significant bits) that determines the meaning of the remaining 10 bits. Here is the complete description of the armlet instruction set, as specified in package armlet:

//
// PROCESSOR INSTRUCTION SET DESCRIPTION
//
// Each instruction is 16 bits in length (possibly followed by 16 bits
// of immediate data), and consists of an opcode and additional information.
// The least significant 6 bits of an instruction specify the 6-bit
// instruction opcode P, followed by additional information depending on
// the layout of the instruction:
//
// 1111110000000000
// 5432109876543210  (bit index)
//
// ????AAALLLPPPPPP  LA:   monadic operator                $L = op $A
// ?BBBAAALLLPPPPPP  LAB:  dyadic operator                 $L = $A op $B
// ???????LLLPPPPPP  LI:   monadic operator with imm data  $L = op I
// ????AAALLLPPPPPP  LAI:  dyadic operator with imm data   $L = $A op I
//
// ??????????PPPPPP  N:    niladic control                 op
// ????AAA???PPPPPP  A:    monadic control                 op $A
// ?BBBAAA???PPPPPP  AB:   dyadic control                  $A op $B
// ??????????PPPPPP  I:    monadic control with imm data   op I
// ????AAA???PPPPPP  AI:   dyadic control with imm data    $A op I
//
// Here:
//
// P = 6-bit opcode (0,1,...,63)
// L = lval (3-bit register identifier)
// A = first rval (3-bit register identifier)
// B = second rval (3-bit register identifier)
// ? = [unused]
//
// The opcodes, their operand types, and operand layouts are as follows:
//
val opcodes = immutable.ArraySeq(

    // Instructions that take register operands (if any)

    (0,  "nop" , "N"   , "N"   ), // no operation
    (1,  "mov" , "RR"  , "LA"  ), // rval to lval
    (2,  "and" , "RRR" , "LAB" ), // bitwise AND of rvals to lval
    (3,  "ior" , "RRR" , "LAB" ), // bitwise IOR of rvals to lval
    (4,  "eor" , "RRR" , "LAB" ), // bitwise EOR of rvals to lval
    (5,  "not" , "RR"  , "LA"  ), // bitwise NOT of rval to lval
    (6,  "add" , "RRR" , "LAB" ), // sum of rvals to lval
    (7,  "sub" , "RRR" , "LAB" ), // difference of rvals to lval
    (8,  "neg" , "RR"  , "LA"  ), // negation of rval to lval
    (9,  "lsl" , "RRR" , "LAB" ), // left logical shift of $A by $B
    (10, "lsr" , "RRR" , "LAB" ), // right logical shift of $A by $B
    (11, "asr" , "RRR" , "LAB" ), // right arithmetic shift of $A by $B
    (12, "loa" , "RR"  , "LA"  ), // load from memory word pointed by rval
    (13, "sto" , "RR"  , "LA"  ), // store to memory word pointed by lval
    (14, "cmp" , "RR"  , "AB"  ), // compare vals, result to status
    (15, "jmp" , "R"   , "A"   ), // jump to rval
    (16, "beq" , "R"   , "A"   ), // ... if status is equal
    (17, "bne" , "R"   , "A"   ), // ... if status is not equal
    (18, "bgt" , "R"   , "A"   ), // ... if status is greater than
    (19, "blt" , "R"   , "A"   ), // ... if status is lesser than
    (20, "bge" , "R"   , "A"   ), // ... if status is greater than or equal
    (21, "ble" , "R"   , "A"   ), // ... if status is lesser than or equal
    (22, "bab" , "R"   , "A"   ), // ... if status is above
    (23, "bbw" , "R"   , "A"   ), // ... if status is below
    (24, "bae" , "R"   , "A"   ), // ... if status is above or equal
    (25, "bbe" , "R"   , "A"   ), // ... if status is below or equal

    // Instructions that take immediate data

    (26, "mov" , "RI"  , "LI"  ), // rval to lval
    (27, "and" , "RRI" , "LAI" ), // bitwise AND of rvals to lval
    (28, "ior" , "RRI" , "LAI" ), // bitwise IOR of rvals to lval
    (29, "eor" , "RRI" , "LAI" ), // bitwise EOR of rvals to lval
    (30, "add" , "RRI" , "LAI" ), // sum of rvals to lval
    (31, "sub" , "RRI" , "LAI" ), // difference of rvals to lval
    (32, "lsl" , "RRI" , "LAI" ), // left logical shift of $A by I
    (33, "lsr" , "RRI" , "LAI" ), // right logical shift of $A by I
    (34, "asr" , "RRI" , "LAI" ), // right arithmetic shift of $A by $B
    (35, "cmp" , "RI"  , "AI"  ), // compare rvals, result to status
    (36, "jmp" , "I"   , "I"   ), // jump to rval
    (37, "beq" , "I"   , "I"   ), // ... if status is equal
    (38, "bne" , "I"   , "I"   ), // ... if status is not equal
    (39, "bgt" , "I"   , "I"   ), // ... if status is greater than
    (40, "blt" , "I"   , "I"   ), // ... if status is lesser than
    (41, "bge" , "I"   , "I"   ), // ... if status is greater than or equal
    (42, "ble" , "I"   , "I"   ), // ... if status is lesser than or equal
    (43, "bab" , "I"   , "I"   ), // ... if status is above
    (44, "bbw" , "I"   , "I"   ), // ... if status is below
    (45, "bae" , "I"   , "I"   ), // ... if status is above or equal
    (46, "bbe" , "I"   , "I"   ), // ... if status is below or equal

    // Unused opcodes and the processor halt

    (47, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (48, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (49, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (50, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (51, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (52, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (53, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (54, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (55, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (56, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (57, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (58, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (59, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (60, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (61, "hlt" , "N"   , "N"   ), // [unused opcode -- will halt processor]
    (62, "trp" , "N"   , "N"   ), // processor trap
    (63, "hlt" , "N"   , "N"   )) // processor halt

Example 1. Using the description above, let us figure out the binary representation of the instruction

sub $2, $0, $1

This is an instruction that takes two register operands (rvalues), $0 and $1, and stores the result in a register (lvalue) $2. Thus, the operand type is RRR (three registers) and the binary operand layout is LAB (dyadic operator layout, register lvalue and two register rvalues). Looking at the table of opcodes above, the opcode for a sub with these operands is 7. We thus conclude that the word

0001000010000111

is the binary representation of “sub $2, $0, $1”.

Example 2. Using the description above, let us figure out the binary representation of the instruction

ior $7, $1, 12345

This is an instruction that takes one register operand ($1) and one immediate operand (the decimal number 12345), and stores the result in a register $7. Thus, the operand type is RRI (two registers and an immediate operand) and the binary operand layout is LAI (dyadic operator layout, register lvalue, one register rvalue, one immediate rvalue). Looking at the table of opcodes above, the opcode for an ior with these operands is 28. We thus conclude that the two words

0000001111011100 and 0011000000111001

are the binary representation of “ior $7, $1, 12345”. The first word configures the operation and the second word contains the immediate data (12345 in binary).

Example 3. Let us figure out the meaning of the binary instruction

0110101111001010.

The least significant 6 bits are 001010, that is, the opcode is 10 (in decimal). Looking at the table of opcodes above, this opcode has operand type RRR and the binary operand layout is LAB. Based on the layout, the next least significant 3 bits, 111, indicate that $L is $7. The next least significant 3 bits, 101, indicate that $A is $5. Finally, the next least significant 3 bits, 110 indicate that $B is $6. The most significant bit is unused. Thus, the instruction that we are looking at is

lsl $7, $5, $6

After these examples we are happy to leave the task of mapping between binary and human-readable representations of armlet instructions to Scala code in package armlet.

The instruction decoder unit

Let us now return to completing the design of the data path. In terms of combinational logic, we need to augment the data path with a unit that configures the data path based on the instruction given as control input (including possibly the immediate data). This functionality is implemented in the instruction decoder unit that takes as input the instruction and the immediate data (if any), and gives as output the control input to the subunits of the data path.

_images/datapath-control.png

Instructing the data path with Trigger

We are now ready for our second Trigger session, which this time includes the armlet data path and the instruction decoder unit.

To play with the data path circuit compile and run the file launchDataPathTrigger.scala in the armlet source code, or copy and paste the following to an armlet-aware console:

import armlet._
new DataPathTrigger()

An interface should pop up:

_images/datapathtrigger.png

Observe that the bits of the registers do not accept toggling any more. Our interface to the architecture is now only through the instruction input, which we can present either in the actual binary encoding by toggling the input bits, or in symbolic form through the text field (which Trigger is then kind enough to translate automatically into binary for us, assuming of course that a valid symbolic instruction was entered!).

So let us push our luck, and key in the instruction “mov $5, 12345” into the text field, taking care to press Enter to have the instruction encoded into binary for us by Trigger:

_images/datapathtrigger2.png

Now trigger the clock:

_images/ifbe.png

We observe that 12345 appears in $5 exactly as instructed:

_images/datapathtrigger3.png

You can now instruct the data path as you like. Just key in the instruction and witness it get executed when you trigger the clock. Alternatively, you can key in the instruction and press Shift+Enter, which both encodes the instruction (assuming it is a valid instruction) and triggers the clock. Type in a sequence of instructions and witness sequential logic in operation!

Say, suppose we want to figure out what is \(123+456+789\). The data path will do this for you, just type in the three instructions:

mov $0, 123
add $0, $0, 456
add $0, $0, 789

Or what is the same, in binary:

0000000000011010 0000000001111011
0000000000011110 0000000111001000
0000000000011110 0000001100010101

Of course this is nothing to write home about yet in terms of programmability, since we have to type (or toggle!) the instructions in, one by one, at the Trigger console. Yet we have now in place all the details how this all takes place in binary, in gate-level sequential logic. That is, we have built from first principles a machine that we can instruct to execute individual operations.

Towards programmability

We now know how to maintain state and how to evolve it using sequential logic. We are now very close to a fully programmable computer architecture and hence close to completing our quest for Module I: The mystery of the computer. The only aspects of a fully programmable architecture that are missing are

  1. storing the program (a sequence of instructions) in memory, and

  2. automating the execution of the program with sequential logic, including extending the instruction set with instructions that control this execution.

We will complete these missing pieces in Round 5: A programmable computer.