コンピューター・アーキテクチャ
Computer Architecture

Hennessy and Patterson Appendix A slides!

Outline of This Lecture

• Assembly Programming
• Stages of Instruction Execution
• Pipelining
• Final Thoughts
• Homework

Note: The MIPS simulator is available on SourceForge. The original spim page is still on the web.

You will need vecadd.s, a short MIPS assembly program.

Here are screen shots from spim (Linux), xspim (Linux), PCspim (Windows) and spim and QtSpim (Mac) versions of the tool.

Things to note in the images above, as well as your own execution runs:

• Watch the advance of the PC (program counter) as you step through the program.
• By convention, most of the registers have two names, Rxx and a more mnemonic one indicating the use.
• The floating point registers are either single-precision, or a pair of them (e.g., F0 and F1) are used as one double-precision FP register.
• Notice especially the precision and accuracy of the FP numbers in the output, compared to the values we intended! This is an important fact!

1. Instruction Fetch cycle (IF)
   Fetch the current instruction from memory, using the program counter (PC) as the address, add 4 to the PC, and store the PC (actually, in MIPS, store the tentative new PC into an internal register called NPC, Next PC).
2. Instruction Decode/register fetch cycle (ID)
   Determine which instruction we are holding, fetch the register values (two, always, in this instruction set), compare the two registers and set the EQUAL flag if equal.
3. Execution/effective address cycle (EX)
   Depending on the instruction type:
   
   • Memory reference: The ALU adds the base register and the offset to form the effective address.
   • Register-Register ALU instruction: perform the operation (e.g., addition, multiplication, logic operation) on the register values fetched by the ID stage.
   • Register-Immediate ALU instruction: perform the instruction on the first register read and the immediate value in the instruction.
4. Memory access (MEM)
   If the instruction is a LOAD or a STORE, do the appropriate thing, otherwise do nothing. (In MIPS, update the PC using either NPC or the output of the ALU operation.)
5. Write-Back cycle (WB)
   If the instruction was LOAD, write the value fetched from memory into the matching register; if it was an ALU operation, write the result to the register.

The MIPS Pipeline

Pipeline Hazards

• Structural hazard: when there is only a single resource, such as a single port to access main memory, and two instructions try to use it at the same time, a structural hazard is hit and one must wait on the other.
• Data hazard:Sometimes one instruction will attempt to use the result of an ALU operation (addition, etc.) before the operation is complete. It must wait, sometimes more than one clock cycle.
• Control hazard:In the simplest pipeline implementations, every time a branch occurs, one or more in-progress instructions must be aborted without changing the state of the system, and new instructions must be fetched.

1. Modify vecadd.s to multiply two four-by-four matrices and print the results (行列の掛け算). Call it arraymult.s. vecadd.sを改変して、4次の正方行列をかけ算するプログラムarraymult.sを 作りなさい. Include a printout of the output. Use these arrays (same as above):
   array1: .float 3.14159265, 2.71828183, 1.0, -0.10 1.0, 0.0, 1.0, 0.0 0.0, 1.0, 0.0, 1.0 -1.0, 1.0, -1.0, 1.0
   array2: .float 2.71828183, 1.0, 3.14159265, 1.0 1.0, 0.0, 1.0, 0.0 -1.0, 1.0, -1.0, 1.0 3.0, 2.0, 1.0, 0.0
2. Take your assembly-language matrix multiplication program and count the following:
   1. Floating-point additions actually executed over all loops
   2. Floating-point multiplications actually executed over all loops
   3. Integer additions/subtractions actually executed over all loops
   4. Branches actually executed over all loops
   5. The number of instructions between branch instructions
3. Calculate the ideal throughput for your assembly program, assuming one instruction per clock cycle. How many clock cycles will your program take? How many seconds is that?
4. Find and describe a real-world pipeline. Include:
   1. The number of stages
   2. Functionality of each stage
   3. Interlocking between stages
   4. Any hazards
   5. How balance in execution time is maintained
5. Pipeline hazards equate to arrows flowing right to left on the figure above. Identify the arrows on the diagram above by type and indicate the maximum delay that the hazard can cause.
6. The three pipeline programs we "executed" during class today are linked to below. Calculate the following for each:
   1. The number of instructions that must be executed. Don't forget to account for the loop in program 3. (n.b.: the #-28 in the branch is decimal!)
   2. The number of clock cycles the entire program takes, accounting for data and control hazards.
   3. The average clock cycles per instruction (CPI) for the program.

第7回 11月5日 Lecture 7, November 5: Memory: Caching and Memory Hierarchy

• Follow-up from this lecture:
  • H-P:Appendix A.1 and A.2
  • P-H:
• For next time:
  • H-P: Appendix C.1, C.2 and C.3
  • P-H:

Additional Information

• The pipeline stages
• The sheet for an instruction
• The first pipeline program
• The second pipeline program (data hazard)
• The third pipeline program (control hazard)
• The Wikipedia page on pipelining is decent. (日本語)
• Class top page
• Elsevier's web page for the textbook.
• My web page on system software.

その他

• Rodney Van Meterのホームページ
• 徳田・村井・楠本・中村・高汐・重近・湧川・バンミーター研究プロジェクト