28 Questions
What is the main advantage of SIMD architecture compared to MIMD architecture?
Potentially more energy-efficient
According to Flynn's Taxonomy, which category does a vector architecture fall under?
SIMD
What type of processor is characterized by high-level instructions on vectors?
SIMD vector processor
What is the main difference between SIMD and MIMD architectures?
Number of data operations
Which of the following is a characteristic of a SIMD vector processor?
One instruction can launch many data operations
What is the main purpose of a GPU?
To accelerate graphics rendering
What is the advantage of using vector computers in high-performance computing?
They can achieve parallel speedup
What type of architecture is commonly used in personal mobile devices and servers?
SIMD
Which of the following is not a category in Flynn's Taxonomy?
Fault tolerance
What type of processor is suitable for tasks that require parallel data operations?
SIMD vector processor
What is the primary advantage of vector processors?
They can perform SIMD operations on large datasets
What is the key characteristic of a vector processor's memory system?
It provides interleaved memory access to entire vectors
What is the purpose of the control unit in a vector functional unit?
To detect hazards and schedule operations
What is the key difference between RV64G and RV64V code for the DAXPY operation?
RV64V code uses vector registers, while RV64G code uses scalar registers
What is the main advantage of using vector instructions over scalar instructions for certain types of computations?
They can perform the same operation on multiple data elements simultaneously
What is the name of the benchmark that the SAXPY or DAXPY loop is part of?
Linpack
What is the purpose of the vsetdcfg instruction in the RV64V code for the DAXPY operation?
To enable vector registers for double-precision floating-point operations
What is the key benefit of using vector processors for certain types of computations?
They can perform SIMD operations on large datasets
What is the main difference between a vector processor and a scalar processor?
A vector processor can perform SIMD operations on multiple data elements, while a scalar processor can only perform operations on a single data element
What is the purpose of the vld instruction in the RV64V code for the DAXPY operation?
To load a vector from memory into a vector register
What is the main optimization technique in RV64V vector instructions?
Stalling only for the first element in each vector
What is the benefit of chaining in vector instruction optimization?
Forwarding results from the first FU to the second FU
What is a convoy in the context of vector instruction optimization?
A set of vector instructions that could potentially execute together without structural hazards or RAW hazards
What is the advantage of modern vector computers with multiple parallel pipelines?
Producing two or more results per clock cycle
What type of hazard is eliminated by chaining in vector instruction optimization?
RAW hazard
What is the purpose of convoys in vector instruction optimization?
To execute vector instructions that do not contain structural hazards or RAW hazards
What is the main difference between RV64G and RV64V vector instructions?
RV64G stalls for each vector element, while RV64V stalls for the first element only
What is the main application of vector processors in high-performance computing?
Executing parallel data operations
Study Notes
Vector Computers
- Flynn's Taxonomy: a classification of computer architectures, including:
- SISD (single instruction, single data)
- SIMD (single instruction, multiple data)
- MISD (multiple instructions, single data)
- MIMD (multiple instructions, multiple data)
SIMD vs MIMD
- SIMD architecture:
- Potentially more energy-efficient than MIMD
- More attractive for personal mobile devices and servers
- Programmer thinks sequentially, but achieves parallel speedup through parallel data
- MIMD architecture:
- Needs to fetch one instruction per data operation
- More flexibility
SIMD Vector Processors
- Processors with high-level instructions on vectors
- Example: Y = a × X + Y, where X and Y are vectors of size n, and a is a scalar
Vector Instructions Optimization
- Chaining: forwarding of element-dependent operations
- Convoys: set of vector instructions that can execute together without structural hazards
- Lanes: modern vector computers with multiple parallel pipelines to execute a vector instruction
Common Applications
- Scientific and engineering applications (e.g., simulations, weather forecasts)
- Multimedia applications (e.g., machine learning algorithms)
Main Characteristics
- Loops parallelism exposed through vector instructions
- Memory system adapted to provide memory access to a whole vector instead of each element
- Hardware checks data hazards only once per vector operand
Basic Architecture
- Scalar unit with a common pipeline
- Vector units with multiple parallel pipelines
- Vector registers with multiple elements (e.g., 32 elements, 64 bits/element)
RISC-V Vector Instruction Set Extension (RVV)
- Vector registers with 32 elements, 64 bits/element
- Scalar registers with 31 GPR and 32 FPR
- Vector functional units with fully pipelined operations
Some Vector Instructions
- ADD: add elements of two vectors
- SUB: subtract elements of two vectors
- DIV: divide elements of two vectors
- Load: load vector register from memory
- Store: store vector register into memory
Operation Example
- Vector loop for RV64V: Y = a × X + Y, where X and Y are vectors of size n, and a is a scalar
- RV64V code for DAXPY (double-precision a × X plus Y) is more efficient than RV64G code (8 instructions vs 258)
Learn about Flynn's Taxonomy, a classification of computer architectures, including SISD, SIMD, MISD, and MIMD, and their comparison.
Make Your Own Quizzes and Flashcards
Convert your notes into interactive study material.
Get started for free
