00. "Computer Architecture" coursework 1 question list

"Computer Architecture" coursework 1 question list

1. What is hierarchical structure, system structure, computer composition, computer implementation, Amdahl's law, principle of program locality, heterogeneous multiprocessing, homogeneous multiprocessing?

2. What is the basis of Flynn's classification to classify computers? Which 4 categories can they be divided into?

3. From the perspective of program execution and data processing, what are the levels of parallelism in computer systems from low to high?

4. Give examples to illustrate the relationship between computer system structure, computer composition and computer implementation.

5. A computer with a main frequency of 400MHz executes a standard test program. The instruction types, execution quantities and average clock cycles in the program are shown in the table below. Find the effective CPI, rate, and program execution time for this computer.

6. If the processing speed of a certain function in the computer system is increased by 10 times, but the processing time of this function is only 40% of the running time of the whole system, how much can the performance of the whole system be improved after adopting this enhanced function method?

7. There are 3 components in a computer system that can be improved, and the component speedup ratios of these 3 components are as follows: component 1 speedup ratio = 30 component 2 speedup ratio = 20 component 3 speedup ratio = 10

8. Assuming that there are 3 types of operations in an application program, through improvement, each operation can obtain different performance improvements. The specific data are as follows:

9. Assume that the proportion of floating-point instructions (FP instructions) is 30%, of which the floating-point square root (FPSQR) accounts for 4% of all instructions, the CPI of FP operations is 5, the CPI of FPSQR operations is 20, and the average of other instructions The CPI is 1.25. There are two improvement schemes. The first one is to reduce the CPI of FPSQR operation to 3, and the second one is to reduce the CPI of all FP operations to 3. Try to compare the degree of improvement of the system performance of the two schemes.

1. What is hierarchical structure, system structure, computer composition, computer implementation, Amdahl's law, principle of program locality, heterogeneous multiprocessing, homogeneous multiprocessing?

Answer: multi-level hierarchical structure-according to the order of computer languages ​​from low-level to high-level, the computer system is divided into multi-level cloud substructure according to function, and each layer is characterized by a different language. These levels are in turn: microprogram machine level, machine language (traditional machine level), operating system virtual machine, assembly language virtual machine, high-level language virtual machine, application language virtual machine, etc.
Computer system structure refers to the computer properties seen by machine language programmers, that is, conceptual structure and functional characteristics.
Computer composition refers to the logical implementation of computer system structure, including the composition of data flow and control flow in the physical machine level, as well as the logic design.
Computer implementation refers to the physical implementation of computer components, including the physical structure of processor main memory and other components, the integration and speed of devices, the division and connection of modules, plug-ins, and backplanes, signal transmission, power supply, cooling, and complete machine. assembly technology, etc.
Amdahl's Law——When a component in a system is improved, the improvement of the overall system performance that can be obtained is limited by the percentage of the execution time of the component in the total execution time.
The principle of program locality—the memory addresses accessed during program execution are not randomly distributed, but relatively clustered.
Heterogeneous multiprocessor system——composed of multiple processors of different types and at least responsible for different functions. They process multiple tasks in sequence according to the order of job requirements, using the principle of time overlap, and each completes the specified functional actions. .
Isomorphic multi-processor system - composed of multiple processors of the same type or at least with the same functions, they simultaneously process multiple tasks that can be executed in parallel in the same job.

2. What is the basis of Flynn's classification to classify computers? Which 4 categories can they be divided into?

Answer: The Flynn taxonomy is classified according to the polyploidy of instruction stream and data stream. It divides the computer system structure into single instruction stream single data stream (SISD), single instruction stream multiple data stream (SIMD), multiple instruction stream Single Data Stream (MISD), Multiple Instruction Multiple Data Stream (MIMD).

3. From the perspective of program execution and data processing, what are the levels of parallelism in computer systems from low to high?

Answer: For example, when designing a main memory system, determining the main memory capacity, addressing method, addressing range, etc. belong to the category of computer system structure; determining the main memory cycle, whether to use parallel main memory logically, logic design, etc. belong to the computer composition The category, and the selection of memory chip type, micro-assembly technology, circuit design, etc. belong to the category of computer implementation.
Computer composition is the logical implementation of computer system architecture. A computer implementation is the physical realization that a computer is composed of. A system structure can have multiple components. A composition can have multiple realizations.

4. Give examples to illustrate the relationship between computer system structure, computer composition and computer implementation.

Answer: From the perspective of data processing, the level of parallelism can be divided into:
(1) string bit string: only one bit of a word is processed at a time. This is the most basic serial processing method, and there is no parallelism;
(2) Word string bit union: all bits of a word are processed at the same time, and different words are serialized. Parallelism has begun to appear;
(3) word and bit string: the same bit (called bit slice) of many words is processed at the same time. This method has high parallelism;
(4) Full parallelism: process all or some bits of many words at the same time. This is the highest level of parallelism.
From the perspective of program execution, parallelism levels can be classified from low to high:
(1) Intra-instruction parallelism: parallelism among micro-operations in a single instruction: (2) Instruction-level parallelism: parallel execution of two or two micro-operations (3) Thread-level parallelism
: execute two or more threads in parallel, usually using multiple threads derived from one process as the scheduling unit; (4
) Task-level or process-level parallelism: execute in parallel Two or more processes or tasks (segments), with subroutines or processes as the scheduling unit;
(5) Job or program-level parallelism: parallel execution of two or more jobs or programs.

5. A computer with a main frequency of 400MHz executes a standard test program. The instruction types, execution quantities and average clock cycles in the program are shown in the table below. Find the effective CPI, rate, and program execution time for this computer.

command type	Number of instructions executed	average number of clock cycles
integer	40000	1
data transmission	65000	2
floating point	7000	4
the branch	1000	2
answer:

(1)CPI =(45000X1+75000X2+8000X4+1500X2)/129500=1.776

(2) MIPS rate=f/ CPI =400/1.776 =225.225MIPS

(3) Program execution time = (45000X1+75000X2+8000X4+1500X2)/400=575us

6. If the processing speed of a certain function in the computer system is increased by 10 times, but the processing time of this function is only 40% of the running time of the whole system, how much can the performance of the whole system be improved after adopting this enhanced function method?

Solution: It can be seen from the meaning of the question that the improvement ratio = 40% = 0.4, and the component speedup ratio = 10. According to Amdahl's law:

System speedup ratio=1/((1-0.4)+0.4/10)=1.5625

After adopting this enhanced function method, the performance of the whole system can be improved to 1.5625 times of the original.

7. There are 3 components in a computer system that can be improved, and the component acceleration ratios of these 3 components are as follows:

Part 1 Speedup = 30 Part 2 Speedup = 20 Part 3 Speedup = 10

(1) If the improvement ratios of component 1 and component 2 are 30% respectively, then when the improvement ratio of component 3 is the system speedup ratio can reach 10?

(2) If the improvement ratios of the three components are 30%, 30%, and 20% respectively, and the three components are improved at the same time, what is the proportion of the execution time of the non-accelerable part of the system to the total execution time?

untie:

8. Assuming that there are 3 types of operations in an application program, through improvement, each operation can obtain different performance improvements. The specific data are as follows:

operation type	Number of programs (millions of instructions)	Execution time (cycles) before improvement	Improved execution time (cycles)
operation 1	10	2	1
operation 2	30	20	15
Operation 3	35	10	3
Operation 4	15	4	1

(1) After the improvement, what are the speedup ratios of various operations?
(2) After each type of operation is improved individually, what are the speedup ratios obtained by the program?

(3) What is the speedup ratio of the whole program after the three types of operations are improved at the same time?

untie:

9. Assume that the proportion of floating-point instructions (FP instructions) is 30%, of which the floating-point square root (FPSQR) accounts for 4% of all instructions, the CPI of FP operations is 5, the CPI of FPSQR operations is 20, and the average of other instructions The CPI is 1.25. There are two improvement schemes. The first one is to reduce the CPI of FPSQR operation to 3, and the second one is to reduce the CPI of all FP operations to 3. Try to compare the degree of improvement of the system performance of the two schemes.

untie: