Back to top-level document list
Documents in category: college
| Improving Branch Prediction by Correlating on Data Values |
college |
| topic: |
Proceedings of the 32nd Annual International Symposium on Microarchitecture, November 1999 |
|
Branch predictors typically use combinations of branch PC bits and branch histories to make predictions.Recent improvements in branch predictors have come from reducing the effect of interference, i.e. multiplebranches mapping to the same table entries. In contrast, the branch difference predictor (BDP) uses data values asadditional information to improve the accuracy of conditional branch predictors. The BDP maintains a history ofdifferences between branch source register operands, and feeds these into the prediction process.
An important component of the BDP is a rare event predictor (REP) which reduces learning time and tableinterference. An REP is a cache-like structure designed to store patterns whose predictions differ from the norm.
Initially, ideal interference-free predictors are evaluated to determine how data values improve correla-tion. Next, execution driven simulations of complete designs realize this potential. The BDP reduces themisprediction rate of five SPEC95 integer benchmarks by up to 33% compared to gshare and by up to 15% compared to Bi-Mode predictors. |
| formats: |
Adobe PDF (109.9kB), PostScript (453.6kB)
|
1999-11-01 |
quality 10 |
|
| Using Data Values to Aid Branch Prediction |
college |
| topic: |
My MSEE Research Report |
|
My research was a concerned with gaining an understanding of some program constructs which
create "hard to predict" branches, and figuring out how to use data-value
information to predict branches. I started with a low-level investigation of the
source-level constructs corresponding to the static branches with the most misses. Next, I
investigated the predictability of register values leading up to these branches. I
constructed several predictors for some particular branches in an effort to improve accuracy by
utilizing available program information.
Data-values have the potential to help branch prediction accuracies a great deal - even
more than 20% accuracy improvement over gshare for particular branches - though work
remains to be done to determine an implementable, efficient method for exploiting this
potential.
This research was meant to provide
some basic insight into the problem so that further, more complete, studies could be
done. The purpose of my research was not to be a comprehensive study, but instead a
time-effective investigation of possible heuristics to aid branch-prediction.
Finally, I contributed to a study of an implementable scheme
which used data-values to aid branch prediction.
|
| formats: |
Adobe PDF (560.9kB), PostScript (2739.6kB)
|
1998-12-17 |
quality 9 |
|
| A Linux Implementation of HIP |
college |
| topic: |
ECE750 (Real-Time Computing) Report, Univ. of Wisconsin - Madison |
|
As network traffic becomes more demanding, the overhead of
invoking a hardware interrupt for each packet becomes high. For
regular, frequent, traffic, such as video or audio streaming,
a hybrid interrupt polling scheme (HIP) can be used to decrease
this overhead. When packets arrive regularly and frequently enough,
HIP disables network interrupts and polls the network interface card (NIC)
during regular kernel context-switches (1ms resolution). Less overhead
allows higher application CPU utilization and higher network bandwidth.
We have implemented HIP in the Linux 2.0.35 kernel for the 3Com 3c590 "Vortex"
NIC (10BaseT), and have collected rudimentary performance information. For high
packet traffic, HIP can reduce overhead up to 20%. HIP increases
packet latency except when packets are very regular, and throughput
for a normal application such as FTP is increased by 8.7%.
|
| formats: |
Adobe PDF (336.2kB), PostScript (657.1kB)
|
1998-12-15 |
quality 5 |
|
| Effectiveness of Arithmetic Code Protection in Dual ALU System |
college |
| topic: |
ECE753 (Fault-Tolerant Computer Systems), Univ. of Wisconsin - Madison |
|
A duplicate system has two identical pieces of hardware
which perform the same computation in parallel and then compare the results; if they disagree, an error message is generated. The duplicate system cannot tolerate faults but can only detect them. There is no method of determining which unit
is faulty. However, a duplicate system can mask faults if the voter knows which unit is faulty.
We implemented an AN code protected duplicate ALU system. We injected arbitrary faults and measured the fault detection capability. We found a 90% improvement in our duplex system over the simplex system. |
| formats: |
Adobe PDF (59.8kB), PostScript (347.7kB)
|
1998-05-09 |
quality 5 |
|
| Study of Relative Performance Impact: Coherence Protocol vs. Network Speed |
college |
| topic: |
ECE757 (Multiprocessor Computer Architecture), Univ. of Wisconsin - Madison |
|
Architects and implementors must know design trade-offs to
design successful systems. Two important aspects of
multiprocessor performance are network quality and
coherence protocol. We study the performance impact of
coherence protocol choice (MSI vs. MESI) as compared to the
performance impact of the high-speed network. A higher
quality network is either wider, or has less latency, or both.
We find that in the majority of cases, network quality
dominates the effects of coherence protocol - that is,
lower-quality networks with the MESI protocol do not yield
faster runs that higher-quality networks with the MSI
protocol. We also note that program characteristics can
affect which protocol performs better. |
| formats: |
Adobe PDF (158.9kB), PostScript (233.6kB), TeX (11.5kB)
|
1998-05-01 |
quality 5 |
|
| Dynamic Instruction Reuse - SPECfp92 |
college |
| topic: |
CS752 (Uniprocessor Architecture), Univ. of Wisconsin - Madison |
|
This report further studies the phenomenon of dynamic instruction reuse as proposed by Sodani and Sohi [1]. We analyze
the behavior of SPECfp92 benchmarks to complement the SPECint92 results in [1]. Our results show that, like integer programs, floating-point programs can gain significant speedup from a reuse buffer, but reuse of floating-point instructions does not contribute a significant amount to that speedup. |
| formats: |
Adobe PDF (131.8kB), PostScript (323.1kB)
|
1997-12-04 |
quality 5 |
|
| A Paper on Engineering Ethics |
college |
| topic: |
ECE350 (Human Communication for EEs) |
|
Engineering is not only about using brick and metal to build physical towers: the
basis of engineering is the ideas that engineers have. Just as burning and
looting buildings is immoral, so is stealing the ideas of another and calling them
your own.
In this paper, we will discuss a hypothetical case developed in Intellectual
Property Rights: A Student's Guide, by Chin Yee Ng at MIT. |
| formats: |
Adobe PDF (67.2kB), PostScript (57.4kB), TeX (10.1kB)
|
1997-03-21 |
quality 5 |
|
| A Brief History of the Web |
college |
| topic: |
ECE350 (Human Communication for EEs) |
|
The World Wide Web (WWW) has become a ubiquitous part of modern
culture. Since its introduction several years ago as the first
easy-to-use network interface, it has rapidly become one of the
most popular ways to retrieve information while creating a new
mode of publishing to the masses.
|
| formats: |
Adobe PDF (42.6kB), PostScript (34.8kB), TeX (7.1kB)
|
1997-02-06 |
quality 5 |
|
| RRisc Architecture Report |
college |
| topic: |
ECE554 (Digital Engineering Laboratory) |
|
The mission of UW--Madison's ECE 554 students is to Design a
non--trivial computer with an original instruction set. This document
describes the superscalar architecture developed by the RRisc team and
how it was implemented in Xilinx XC4000 FPGAs using the WICEPS
prototype board.
|
| formats: |
Adobe PDF (397.7kB), PostScript (301.3kB), TeX (2.3kB)
|
1996-12-17 |
quality 5 |
|
| Distributive Justice: A King's Castle Is His Home |
college |
| topic: |
Philosophy - Moral Issues (341), Univ. of Wisconsin - Madison |
|
Should the strong be required to support the weak? How
does a society "distribute" wealth among its members? These
are the questions of Distributive Justice. There are three
basic sides to this issue. The Permissive system entitles
individuals to a subsistence income simply for existing as a
human. The Puritan system requires that people at least be
willing to contribute to society in order to receive a
subsistence income. Finally, the Individualist view holds
the property rights of the individual to be sacred: no one
may forcibly deprive him of his goods
I will argue for the last alternative. |
| formats: |
HTML (8.8kB), Adobe PDF (58.3kB), PostScript (108.3kB), TeX (9.1kB)
|
1995-04-26 |
quality 5 |
|
| Final Report on Test Generation |
college |
| topic: |
ECE553 (Testing and Testability of Digital Systems) |
|
Problem Statement:
1. Given a circuit description, generate a compact test set with high fault coverage.
2. Using test set from (1), determine whether three supplied circuits contain
an injected fault.
3. Finally, determine the fault location in a fourth supplied circuit.
|
| formats: |
Adobe PDF (151.6kB), PostScript (394.8kB), TeX (1.1kB)
|
1997-11-30 |
quality 4 |
|
| Angst Overflow |
college |
| topic: |
What happens when a hacker has too much free time in grad. school |
| formats: |
HTML (5.0kB)
|
1997-10-26 |
quality 4 |
|
| Short version of RRisc Architecture Report |
college |
| topic: |
ECE350 (Human Communication for EEs) |
|
During the Fall semester of 1996, the RRisc team designed,
debugged, and built a "non--trivial computer with an original instruction set," as specified in
the project requirements for ECE554 at the University of
Wisconsin - Madison.
After specification and design, the computer was implemented in
hardware using Xilinx XC4000 series FPGAs, with support circuitry
on the WICEPS board supplied for the course.
The main feature of our computer was that it was super--scalar; it could
execute up to 3 instructions simultaneously.
The project was a success: once in hardware and debugged, the
system ran at 4.09 MHz, a period of 209ns.
|
| formats: |
Adobe PDF (191.1kB), PostScript (520.4kB), TeX (1.5kB)
|
1997-05-01 |
quality 4 |
|
| Dynamic Programming: All-Points Shortest-Path Problem |
college |
| topic: |
CS577 (Algorithms) |
|
Dynamic programming finds optimal solutions to problems by recursively figuring out
which sub-solution, combined with the choices at the current level, will produce
the best result |
| formats: |
Adobe PDF (45.7kB), PostScript (31.6kB), TeX (3.7kB)
|
1997-04-28 |
quality 4 |
|
| Minority Influence |
college |
| topic: |
CA266 (Theory of Groups) |
|
Mascovici, Lage, and Naffrechoux (1972) demonstrated that a constant, repetitive, vocal minority can
influence the majority. Nemeth, Swedlund, and Kanki built on this to show that repetition is not necessary; it is
only necessary that the majority perceive confidence and consistency in the viewpoint of the minority for
minority influence to occur. Minority and majority influence is linked to normative and informational
influence; the minority can only have informational influence and the majority uses normative influence
because it defines the group norms. |
| formats: |
HTML (6.4kB)
|
1996-10-01 |
quality 4 |
|
| Communication Networks |
college |
| topic: |
CA266 (Theory of Groups) |
|
The results of the exercise illustrated the differences between autocratic and democratic groups. In summary,
the most important differences are that democratic groups always take more time, always have more satisfied
members, and have accuracy directly proportional to task difficulty. Conversely, autocratic groups take less
time, always have less satisfied members, and have accuracy inversely proportional to task difficulty.
|
| formats: |
HTML (5.6kB)
|
1996-10-01 |
quality 4 |
|
| Group Decision Experiment |
college |
| topic: |
CA266 (Theory of Groups) |
|
On Friday, November 1, my group took part in an experiment that studied various aspects of the group
decision--making process. The situation was this: we were district attornies trying to decide who was the best
person to prosecute in a murder case. We were told that the information we were given might be different, and
we were told that it would be a judgemental task, that there was no "right" answer.
In conclusion, we had already made up our minds before coming into the meeting. Each of us had all the
information and did not have to discuss to gain new information from other group members. The only change
that occured was a clarification of the problem, which changed Andy's mind. |
| formats: |
HTML (5.2kB)
|
1996-10-01 |
quality 4 |
|
| Work Summary Report |
college |
| topic: |
Summer 1996 at HP-Colorado |
|
I have been working at the COL division of Hewlett-Packard, located
in Colorado Springs, Colorado, for the last 2 months. We design,
manufacture, and support logic analyzers, processor
emulators, and other embedded system development tools. This
document describes my experiences by including a general description
of my tasks and projects, skills I developed, observations regarding
the work environment, and the strengths and weaknesses I displayed.
Hewlett-Packard is a very large company, but its divisions are locally
managed, which helps to maintain a small-company feel. |
| formats: |
Adobe PDF (47.0kB), PostScript (85.8kB), TeX (8.8kB)
|
1996-07-25 |
quality 4 |
|
| Investigation into the Construction of Parallel Plate Capacitors |
college |
| topic: |
ECE220 (Electrostatics) |
|
We designed and built two capacitors based on the parallel
plate model: one consisting of two parallel disks separated
by paper; and another consisting of many dielectric
and conducting layers.
We were able to achieve capacitances
of 1.21nF and 61.7nF respectively. Modifying our
dielectric by adding "Inductor Oil" further increased
each roughly by a factor of 2 to 2.15nF and 121.8nF.
Our results confirmed the predicted C = E*A/d. |
| formats: |
Adobe PDF (80.7kB), PostScript (168.9kB), TeX (14.9kB)
|
1995-12-10 |
quality 4 |
|
| Short Paper about Group Darwinism in the Manhatten Project Era |
college |
| topic: |
History of Science in the 20th Century |
|
During WWII, while there was still a global threat by the Stalin
and Hitler, the scientists who could, flocked to the United
States and worked on the Bomb Project. Science could not exist
in an atmosphere that exists under an oppressive dictatorship. |
| formats: |
Adobe PDF (30.2kB), PostScript (54.4kB), TeX (2.8kB)
|
1995-12-06 |
quality 4 |
|
| Short paper relating to Big Science. |
college |
| topic: |
History of Science in the 20th Century |
|
Again, Big Science has pushed forward scientific progress, as it has
with Atomic Physics and others. This time it was George Ellery Hale who
was the great builder. |
| formats: |
Adobe PDF (30.3kB), PostScript (59.7kB), TeX (2.5kB)
|
1995-11-29 |
quality 4 |
|
| Short Paper about Truth in the Manhatten Project Era |
college |
| topic: |
History of Science in the 20th Century |
|
The period of 1937-1940, the years just before the Second World
War, juxtaposed two opposing theories of knowledge and truth.
One was the scientific search for beauty and order in the universe,
as the physicists of the era were doing; the other, proferred by
the politicians and tyrants, was one of hatred and evil. |
| formats: |
Adobe PDF (30.6kB), PostScript (57.2kB), TeX (3.1kB)
|
1995-10-31 |
quality 4 |
|
| Self-evaluation of talk |
college |
| topic: |
ECE350 (Human Communication for EEs) |
|
On March 7, 1997 I gave a talk titled "The Existential Pleasures
of Engineering: The LCD Module via Parallel Port Project." The
audience was a group of engineers, and the purpose of the talk
was to give a technical overview of the project. |
| formats: |
Adobe PDF (36.7kB), PostScript (92.9kB), TeX (3.1kB)
|
1997-05-01 |
quality 3 |
|
| Report on a talk by Miguel Lorne titled Garveism and the Rastafarian Movement in the Caribbean and their Implication on the Americas. |
college |
| topic: |
ECE350 (Human Communication for EEs) |
|
On Thursday, February 13, I attended a talk given by Miguel Lorne
titled, "Garveism and the Rastafarian Movement in the Caribbean
and their Implication on the Americas."
Once I got past the strangeness of the situation, I realized that
Miguel was a good speaker, for the most part. This was not a technical
talk, but a time to for him to tell stories about himself and the
history of people in the Caribbean. |
| formats: |
Adobe PDF (33.2kB), PostScript (27.2kB), TeX (3.2kB)
|
1997-05-01 |
quality 3 |
|
| Is Arrowhead High School trying to censor my brother? |
college |
| topic: |
Regarding Abe's ANTI-AHS Web Page |
| formats: |
HTML (1.3kB)
|
1997-03-10 |
quality 3 |
|
| On Leadership Attributes |
college |
| topic: |
CA266 (Theory of Groups) |
|
Every successful leader must have a basic set of certain traits, however the role of the leader in the group
determines which extra, specialized traits that leader must have. I will discuss how our group's list of leadership
traits, and those of our leader, compare to those special characteristics required of military, corporate, sports,
and academic leaders. |
| formats: |
HTML (5.6kB)
|
1996-09-01 |
quality 3 |
|
| Short paper on the Manhatten Project |
college |
| topic: |
History of Science in the 20th Century |
|
The scientists frantically worked on the Manhatten Project so
that the world would not be overtaken by Germany's facist
rulers. That was a moral decision on their part to
keep something they valued - a more free world, one
on which they wanted to live. That was their moral justification
of the production of the bomb.
|
| formats: |
Adobe PDF (26.4kB), PostScript (48.5kB), TeX (1.8kB)
|
1995-11-13 |
quality 3 |
|
| A short paper relating to the Division of Scientists during the Manhatten Project Era. |
college |
| topic: |
History of Science in the 20th Century |
|
While the world was dividing just before the Second World War,
the international physicists were trying to stay together. They
were torn by pressures from Gorvernments who wanted power. |
| formats: |
Adobe PDF (25.5kB), PostScript (46.3kB), TeX (1.8kB)
|
1995-11-08 |
quality 3 |
|
| Paper comparing and contrasting Watson and Jungk's versions of Science. |
college |
| topic: |
History of Science in the 20th Century |
|
Watson's The Double Helix and Jungk's Brighter Than a Thousand Suns give two sharply contrasting
views of science, and the behavior of scientists. Many think
Watson's is a "dirty" science, while Jungk's is "pure, clean"
science. Just as it is an error to stereotype all
people who live in Wisconsin as farmers, so it is an error
to assume all scientists are the same.
|
| formats: |
Adobe PDF (39.9kB), PostScript (77.6kB), TeX (3.3kB)
|
1995-11-08 |
quality 3 |
|
| What drives me to do engineering |
college |
| topic: |
Trying to get an enginering scholarship |
|
I am driven to apply technology to find excellent solutions.
Making things work in elegant ways is something that excites and motivates me. There is a unique pleasure in
creating a device with an elegant design and watching it work correctly, efficiently. |
| formats: |
HTML (4.8kB)
|
1996-03-21 |
quality 2 |
|
| Why I do engineering, II. |
college |
| topic: |
A scholarship application |
|
The difference between what is possible and what is commonly available frustrtes me. I want to apply
technology to find excellent solutions. |
| formats: |
HTML (2.2kB)
|
1995-01-01 |
quality 2 |
|
| My work experience, honors and awards, etc |
college |
| topic: |
Trying to get an engineering scholarship. |
| formats: |
HTML (3.9kB)
|
1996-03-21 |
quality 0 |
|
|