Chapter 2
Data Level Parallelism
Book 1 – Computer Architecture: A Quantitative Approach, Henessy and Patterson,
5th Edition, Morgan Kaufmann, 2012
Chapter 4 - Data-Level Parallelism in Vector, SIMD, and GPU Architectures
Parallelism
Classes of parallelism in applications
• Data Level Parallelism
oMany data items can be operated on at the same time
• Task Level Parallelism
oDifferent tasks are created that can operate independently and
largely in parallel.
Flynn’s Taxonomy
SIMD
True SIMD
One CPU (control Unit) + multiple
ALUs (Processing Elements(PEs))
each with a memory (can be shared
memory)
Pipelined SIMD
One CPU (control Unit) + pipelined
ALU
ALU work in a pipelined manner not
independently
Data Level Parallelism - Single Instruction stream,
Multiple Data Stream (SIMD)
Three variants
• Vector Architectures
• Multimedia SIMD Extensions
• GPUs, APUs
Vector Processing
• Vector – a set of scalar data elements, all of the same type, stored in memory
• Vector Processor – an ensemble of hardware resources, including vector
registers, functional pipelines, processing elements, and register counters for
performing vector operations
• Vector Processing occurs when arithmetic and logical operations are applied to
vectors
Properties of Vector Processors
• Vector Operations : arithmetic (add, sub, mul, div), memory accesses,
effective address calculations
• Multiple vector instructions can be in progress at the same time =>
more parallelism
• Applications to benefit
• Large scientific and engineering applications (simulations, weather
forecasting, applications involving large matrix operations)
• Multimedia applications
(video codecs, image processing, audio processing)
Basic Vector Architectures
• Vector Processor : ordinary pipelined scalar unit + vector unit
• Types of vector processors
• Memory-Memory processors: all vector operations are memory to
memory (CDC)
• Vector-Register processors: all vector operations except load and
store are among the vector registers (CRAY-1, CRAY-2, X-MP, Y-MP)
➢VMIPS – Vector processor as an extension of the 5-stage MIPS processor
Components of VMIPS Processor
• Vector registers—Each vector register is a fixed-length bank holding a single
vector
➢Vector register has at least 2 read and 1 write port
➢Typically 8-32 vector registers, each holding 64-128, 64 bit elements
➢VMIPS - 8 vector registers, each holding 64 elements of 64 bits (16 Rd ports, 8
Wr ports)
• Vector Functional Units (FUs) : fully pipelined, can start new operation every
clock cycle
• Typically 4 to 8 FUs: FP add, FP mult, FP reciprocal, integer add, logical, shift
• May have multiple of same unit
• VMIPS : 5 FUs (FP add/sub, FP mul, FP div, integer, FP logical)
Components of VMIPS Processor
• Vector Load-Store Units (LSUs)
➢Fully pipelined
➢May have multiple LSUs
➢VMIPS – 1 VLSU, bandwidth is 1 word per cycle after initial delay
• Scalar Registers
➢Single element for FP scalar or address
➢VMIPS – 32 GPR, 32 FPRs they are read out and latched at one input of the
FUs
• Cross-bar to connect FUs, LSUs, registers
