Polymorphous Architectures: A Unified Approach for Extracting Concurrency of Different Granularities
| Sorted by Date | Classified by Publication Type | Classified by Project |
Karthikeyan Sankaralingam. Polymorphous Architectures: A Unified Approach for Extracting Concurrency of Different Granularities. Ph.D. Thesis, The University of Texas at Austin, Department of Computer Sciences, 2006.
Download
Abstract
Processor architects today are faced by two daunting challenges:emerging applications with heterogeneous computation needs andtechnology limitations of power, wire delay, and processvariation. Designing multiple application-specific processors orspecialized architectures introduces design complexity, a softwareprogrammability problem, and reduces economies of scale. There is apressing need for design methodologies that can provide support forheterogeneous applications, combat processor complexity, and achieveeconomies of scale. In this dissertation, we introduce the notion ofarchitectural polymorphism to build such scalable processors thatprovide support for heterogeneous computation by supporting differentgranularities of parallelism. Polymorphism configures coarse-grainedmicroarchitecture blocks to provide an adaptive and flexible processorsubstrate. Technology scalability is achieved by designing anarchitecture using scalable and modular microarchitecture blocks.We use the dataflow graph as the unifying abstraction layer acrossthree granularities of parallelism--instruction-level, thread-level,and data-level. To first order, this granularity of parallelism is themain difference between different classes of applications. Allprograms are expressed in terms of dataflow graphs and directly mappedto the hardware, appropriately partitioned as required by thegranularity of parallelism. We introduce Explicit Data Graph Execution(EDGE) ISAs, a class of ISAs as an architectural solution forefficiently expressing parallelism for building technology scalablearchitectures.We developed the TRIPS architecture implementating an EDGE ISA using aheavily partitioned and distributed microarchitecture to achievetechnology scalability. The two most significant features of the TRIPSmicroarchitecture are its heavily partitioned and modular design, andthe use of microarchitecture networks for communication acrossmodules. We have also built a prototype TRIPS chip in 130nm ASICtechnology composed of two processor cores and a distributed 1MBNon-Uniform Cache Access Architecture (NUCA) on-chip memory system.Our performance results show that the TRIPS microarchitecture whichprovides a 16-issue machine with a 1024-entry instruction window cansustain good instruction-level parallelism. On a set of hand-optimizedkernels IPCs in the range of 4 to 6 are seen, and on a set ofbenchmarks with ample data-level parallelism (DLP), compiler generatedcode produces IPCs in the range of 1 to 4. On the EEMBC and SPECCPU2000 benchmarks we see IPCs in the range of 0.5 to 2.3. Comparingperformance to the Alpha 21264, which is a high performancearchitecture tuned for ILP, TRIPS is up to 3.4 times better on thehand optimized kernels. However, compiler generated binaries for theDLP, EEMBC, and SPEC CPU2000 benchmarks perform worse on TRIPScompared to an Alpha 21264. With more aggressive compiler optimizationwe expect the performance of the compiler produced binaries toimprove.The polymorphous mechanisms proposed in this dissertation areeffective at exploiting thread-level parallelism and data-levelparallelism. When executing four threads on a single processor,significantly high levels of processor utilization are seen; IPCs arein the range of 0.7 to 3.9 for an application mix consisting of EEMBCand SPEC CPU2000 workloads. When executing programs with DLP, thepolymorphous mechanisms we propose provide harmonic mean speedups of2.1X across a set of DLP workloads, compared to an execution model ofextracting only ILP. Compared to specialized architectures, thesemechanisms provide competitive performance using a single executionsubstrate.
Additional Information
This is a test of the extra info broadcasting system.
BibTeX
@phdthesis{diss06:sankaralingam,
author={Karthikeyan Sankaralingam},
title="{Polymorphous Architectures: A Unified Approach for Extracting Concurrency of Different Granularities}",
school={The University of Texas at Austin, Department of Computer Sciences},
year={2006},
month={October},
abstract = {
Processor architects today are faced by two daunting challenges:
emerging applications with heterogeneous computation needs and
technology limitations of power, wire delay, and process
variation. Designing multiple application-specific processors or
specialized architectures introduces design complexity, a software
programmability problem, and reduces economies of scale. There is a
pressing need for design methodologies that can provide support for
heterogeneous applications, combat processor complexity, and achieve
economies of scale. In this dissertation, we introduce the notion of
architectural polymorphism to build such scalable processors that
provide support for heterogeneous computation by supporting different
granularities of parallelism. Polymorphism configures coarse-grained
microarchitecture blocks to provide an adaptive and flexible processor
substrate. Technology scalability is achieved by designing an
architecture using scalable and modular microarchitecture blocks.
We use the dataflow graph as the unifying abstraction layer across
three granularities of parallelism--instruction-level, thread-level,
and data-level. To first order, this granularity of parallelism is the
main difference between different classes of applications. All
programs are expressed in terms of dataflow graphs and directly mapped
to the hardware, appropriately partitioned as required by the
granularity of parallelism. We introduce Explicit Data Graph Execution
(EDGE) ISAs, a class of ISAs as an architectural solution for
efficiently expressing parallelism for building technology scalable
architectures.
We developed the TRIPS architecture implementating an EDGE ISA using a
heavily partitioned and distributed microarchitecture to achieve
technology scalability. The two most significant features of the TRIPS
microarchitecture are its heavily partitioned and modular design, and
the use of microarchitecture networks for communication across
modules. We have also built a prototype TRIPS chip in 130nm ASIC
technology composed of two processor cores and a distributed 1MB
Non-Uniform Cache Access Architecture (NUCA) on-chip memory system.
Our performance results show that the TRIPS microarchitecture which
provides a 16-issue machine with a 1024-entry instruction window can
sustain good instruction-level parallelism. On a set of hand-optimized
kernels IPCs in the range of 4 to 6 are seen, and on a set of
benchmarks with ample data-level parallelism (DLP), compiler generated
code produces IPCs in the range of 1 to 4. On the EEMBC and SPEC
CPU2000 benchmarks we see IPCs in the range of 0.5 to 2.3. Comparing
performance to the Alpha 21264, which is a high performance
architecture tuned for ILP, TRIPS is up to 3.4 times better on the
hand optimized kernels. However, compiler generated binaries for the
DLP, EEMBC, and SPEC CPU2000 benchmarks perform worse on TRIPS
compared to an Alpha 21264. With more aggressive compiler optimization
we expect the performance of the compiler produced binaries to
improve.
The polymorphous mechanisms proposed in this dissertation are
effective at exploiting thread-level parallelism and data-level
parallelism. When executing four threads on a single processor,
significantly high levels of processor utilization are seen; IPCs are
in the range of 0.7 to 3.9 for an application mix consisting of EEMBC
and SPEC CPU2000 workloads. When executing programs with DLP, the
polymorphous mechanisms we propose provide harmonic mean speedups of
2.1X across a set of DLP workloads, compared to an execution model of
extracting only ILP. Compared to specialized architectures, these
mechanisms provide competitive performance using a single execution
substrate.
},
bib_dl_pdf = "http://www.cs.wisc.edu/~karu/docs/diss/karu-diss.pdf",
bib_dl = "http://www.cs.wisc.edu/~karu/docs/diss/",
bib_pubtype = {Other},
bib_rescat = {proj-trips},
bib_extra_info = {This is a test of the extra info broadcasting system.}
}
Generated by bib.pl (written by Patrick Riley ) on Sun Sep 26, 2021 16:14:28 time=1207019082