Architecture

Assignment: Architecture

• Assign: Monday 3 October
• Due: Thursday 13 October
• Policy: Individual graded synthesis assignment
• Submit: Upload a PDF written on the cs240-arch-worksheet.pdf submission sheet
• Reference:
  • Sequential Logic
  • A Simple Processor

Contents

• Assignment: Architecture
• Exercises
  • 1. Flop-flip-flopping (10 points)
  • 2. Reconstructing Memories (14 points)
  • 3. A Loopy Program (15 points)
  • 4. Taking Control (14 points)
  • 5. Jumping into the Unknown (15 points)
  • 6. Instruction Not Missing (12 points)
  • 7 Extra Fun Points Affixed or Afloat in a C of Numbers
• Submission

Exercises

Please write your answers on the cs240-arch-worksheet.pdf submission sheet to streamline the grading process. Submit your work by scanning a PDF of all worksheet pages (in order) and uploading it to Gradescope. Remember that this course uses your Gradescope account attached to your Wellesley email address (the variety that matches your username). You can merge accounts across multiple email addresses if helpful.

On Tue Oct 11, Lyn made minor changes to the worksheet to make it easier to grade in Gradescope. However, if you started the assignment with the old worksheet, it’s fine to use that and you should not use the new worksheet.

1. Flop-flip-flopping (10 points)

Consider the following circuit that uses falling-edge triggered D flip-flops.

Assume the Clock input, and outputs Q₀, Q₁, and Q₂, are all initially 0.
In one Clock cycle, the Clock is first at level 0 for a duration d, then has a rising edge that jumps up to level 1 for the same duration d, and then ends with a falling edge that jumps back down to 0. Draw waveforms for Clock, Q₀, Q₁, and Q₂, showing at least ten Clock cycles. You do not need to turn in your waveform drawing, but it will help you reason about the circuit.

1. Write a table summarizing the values of Q₀, Q₁, and Q₂ after each Clock cycle.

2. Briefly explain what this circuit does at a high level, treating the Q outputs together. A phrase to a couple sentences is enough.

2. Reconstructing Memories (14 points)

In this set of exercises, you will reuse provided RAMs (random access memories) to design RAMs of different dimensions without reimplementing the internals.

Provided RAMs

In an A×B RAM, A is the number of addressable locations (cells) in the memory and B is the width of each location (cell) in bits. Recall that a RAM is much like a register file, but the underlying storage technology is not based on flip-flops. Memory addresses are analogous to register numbers in a register file; memory locations (cells) are analogous to registers in a register file. Unlike the register file we designed, these RAMs support reading only one location at a time.

The provided RAMs have the following form:

• Each RAM has a Data In port, a Data Out port, an Address input and a Write Enable input.
• Each RAM is always reading from the location given by the current Address, expressing its contents on Data Out.
• Each RAM also stores the value on Data In into the location given by the current Address if the Write Enable input carries a 1.
• Each RAM is indivisible.

Rules

You will design new RAMs with the same kind of inputs, outputs, and externally observable behavior, but with different dimensions. For each:

• Draw one or more small boxes of the form above, representing the provided RAM(s), inside a larger box of the same form, representing the RAM you are implementing.
• Connect inputs and outputs of the larger box with those of the smaller boxes, possibly adding combinational logic building blocks.
• Do not draw each individual line of a bus (e.g., Address) unless necessary to distinguish the endpoints of the individual lines. Use hatch marks and width annotations instead. You must specify the number of bits (width of bus) carried by each bus. When a larger bus splits into or is joined from smaller buses, you must show which bit ranges from the larger bus correspond to the width of the smaller bus, as shown in the example below:

Exercises

1. (4 points) Design a 256×8 RAM by adding minimal wiring and logic to two identical 256×4 RAM components. Label the widths (in bits/lines) of the ports Address, Data In, and Data Out for each 256×4 RAM you use and for the 256×8 RAM you implement.

2. (10 points) Design a 64K×8 RAM by adding minimal wiring and logic to a single provided 16K×32 RAM. (K = “kilo” = 1024)

   • Unlike the previous exercise, this provided RAM component has the same capacity as the desired final product, though it has different dimensions:
     • The provided RAM has fewer addressable memory locations (A) than the RAM you will design.
     • The provided RAM has a larger addressable unit of data (B) than the RAM you will design.
   • Be careful with writes: when writing an 8-bit value into the provided RAM, you must be sure to change only the relevant 8 bits of storage. All other storage bits should hold the same values as before the write operation.

3. A Loopy Program (15 points)

In this exercise you will simulate the execution of an HW ISA program with a loop.

Background

In the lectures on the HW ISA architecure, there are examples of using an execution table to simulate the execution of HW ISA programs, one of which contains a loop.

Here we consider a different program that has a loop to remind you of the conventions used in an execution table. Suppose that register R2 has been initialized with the signed decimal number 3 and register R3 has been initialized with the signed decimal number 7. Also suppose that the program P0 has the following Instruction Memory (IM):

Then here is an execution table that shows the step-by-step execution of the program P0:

Your Task

This task considers a HW ISA program P1 with the following Instruction Memory IM:

1. (10 points) In the worksheet, create an execution table that shows the execution of program P1. Use the same notational conventions used in the example execution table for P0 above. Any numbers beginning with 0x will be interpreted as hexidecimal; any numbers not beginning with 0x will be interpreted as decimal.

2. (1 point) Show the final values of the registers R2, R3, and R4 when the program execution halts. Again, any numbers beginning with 0x will be interpreted as hexidecimal; any numbers not beginning with 0x will be interpreted as decimal.

3. (4 points) Translate program P1 (except for the final HALT instruction) into a sequence of statements in one of Python, Java, or JavaScript (you choose!), where one of the statements is a while loop. Your program should included assignments to variables named R2, R3, and R4 that represent the registers with these names in P1. You may assume that the variables R0 and R1 have been initialized to 0 and 1, respectively, before the statements you write.

4. Taking Control (14 points)

In this exercise you will design the Control Unit for the HW microarchitecture we designed in class. Our design focused on connecting the datapath components (e.g., the register file, ALU, memory, etc.), but we left the Control Unit as black box.

Give a truth table with 4-bit HW instruction opcodes as inputs, plus one output for each of the control lines, outputs of the Control Unit that are control inputs to other components. The table should have one row for each of the seven instructions (one row for each opcode) except JMP from the first instruction table here.) The control lines we need are:

• Reg Write (1 bit): controls the write-enable of the register file.
• ALU Op (4 bits): controls the ALU (with the same 4 ALU control lines we developed earlier).
  • From left to right: Invert A, Negate B, Op₁, Op₀.
• Mem Store (1 bit): controls the write-enable of the data memory.
• Mem (1 bit): controls three multiplexers deciding which input to provide for:
  • the register file’s write address;
  • the register file’s write data; and
  • the ALU’s second operand.
• Branch (1 bit, added during class): controls whether to choose the next PC based on a branch target and test

Please note that Mem Store, Mem, and Branch are all active high (0 = False, 1 = True).

5. Jumping into the Unknown (15 points)

1. (9 points) On Tue Oct 11, Lyn made minor changes to the wording of this problem part to match the updated worksheet. The essence of the problem has not changed; it has just been reorganizezd to clarify the three subparts and where their answers should be entered on the worksheet.

   Add logic to the HW microarchitecture to implement the JMP instruction, which we did not implement in class. JMP sets the PC to the absolute instruction address given by its argument (a number) multiplied by 2. For example, JMP 3 sets the PC to 0x6, causing the instruction stored at address 0x6 (i.e., the 3rd instruction in the program) to be executed next.

   1. (6 points) In the wiring diagram for the microarchitecture on the worksheet, you will add:

      • One new Jump control line from the Control Unit (in Problem 4), carrying 1 if the current instruction is JMP and 0 otherwise.

      • Wiring and logic using the Jump control line and the offset bits from the JMP instruction encoding to store the JMP instructions’s target address to the PC if and only if the current instruction is JMP. You will need to “cut” one existing wire to splice in some new logic.

   2. (1 point) In the Control Unit Truth Table from the previous problem, flesh out the contents of the the Jump control line in the final column.

   3. (2 points) In the Control Unit Truth Table from the previous problem, flesh out the contents of the row for the JMP opcode.

2. (6 points) Assume R2 and R3 hold input values when the following program starts. R4 will hold an output when the program stops. The address of each instruction in the instruction memory is shown at left. The (new) HALT instruction stops the computer.

   1. (4 points) Execute the program assuming R2 holds 5 and R3 holds 2. What value does R4 hold when the computer reaches HALT?
   2. (2 points) Try a couple more examples with (small) positive numbers in R2 and R3. What does this code do with R2 and R3 to compute R4? Answer with one simple line of valid C code of the form: R4 = ...;. The C could use variables R2, R3, and R4 (corresponding to the HW registers) and any simple C operations that accomplsih the same effect as these combined HW instructions. C control flow is not needed.

    # R2 and R3 hold program inputs here.
0:  AND R2, R2, R4
2:  AND R3, R3, R5
4:  BEQ R5, R0, 3
6:  SUB R5, R1, R5
8:  ADD R4, R4, R4
A:  JMP 2
C:  HALT # Stops execution.
    # What value is in R4?

6. Instruction Not Missing (12 points)

The HW ISA introduced in class lacks an instruction for bitwise complement. A logical choice would be NOT Rs,Rd, which takes the bitwise complement of the value in source register Rs and stores into destination register Rd. You will design two implementations of the proposed NOT instruction.

1. (4 points) One way to implement NOT is to replace any occurence of a NOT instruction by a sequence of existing HW instructions that has the same effect as NOT.

   Given source register identifier Rs and destination register identifier Rd, write a sequence of one or more HW instructions that has the same effect as NOT Rs,Rd. The replacement instructions may use values from any registers, but must store values only in Rd. Instructions may not store values in any other register or data memory location. The sequence may use only the instructions defined in class. (See the instruction table here.)

   Hint: Think about the relation of bitwise complement and arithmetic rules.

2. (4 points) A second way to implement NOT is to add a new instruction to the ISA, complete with an unique encoding, and to extend the microarchitecture to implement it. First, implement a related instruction (NAND) in this way.

   1. Define a 16-bit encoding of the NAND instruction as in the instruction table here. Choose any unused opcode for the encoding.
   2. Add a row to the Control Unit Table from the previous problem to indicate how to control the datapath to implement NAND. Do not add extra logic to the datapath.

3. (4 points) Consider how to implement NOT using NAND. Define an encoding for the NOT instruction directly in the ISA. Choose the encoding carefully to minimize additions to the Control Unit and the datapath. Assuming you have completed (b), an ideal encoding requires no new rows or columns in the Control Unit table.

   Hint: Think of how to translate the NOT instruction to a NAND instruction and use this intuition when designing the bit encoding of NOT.

7 Extra Fun Points Affixed or Afloat in a C of Numbers

1. The Sea language (an imaginary programming language otherwise identical to C) supports a signed fixed8ths char type¹ using an 8-bit signed fixed point number system with precision to the 1/8 (eighths) place.

   1. What are the minimum and maximum numbers representable by a signed fixed8ths char? Write your answer as a base ten number using fractions or decimals.

   2. What are the minimum and maximum numbers representable by a signed fixed32nds char, which uses 8-bit signed fixed-point representation with precision to the 32nds place? Write your answer as a base ten number using fractions or decimals.

   3. 

      We provide a hardware adder design (at right) that supports 8-bit two’s complement addition (e.g., for signed char numbers). Draw a logic diagram for a hardware adder that supports the signed fixed8ths char number system. You may use the provided two’s complement adder as a black box (no need to draw the internals) and extend it with whatever wiring and logic is necessary. There is a “trick” here! Keep it simple.

2. Using the 6-bit floating-point encoding defined in class, give decimal notation of the numbers represented by the following floating-point encodings.

   1. 110101
   2. 100001
   3. 011100
   4. 000011
   5. 010010
   6. 111101

Submission

Please write your answers on the cs240-arch-worksheet.pdf submission sheet to streamline the grading process. Submit your work by scanning a PDF of all worksheet pages (in order) and uploading it to Gradescope. Remember that this course uses your Gradescope account attached to your Wellesley email address (the variety that matches your username). You can merge accounts across multiple email addresses if helpful.