data编程辅导 、讲解 C/C++，Python程序
Design Space Exploration
Solutions to project assignments are to be developed within your group, without col laboration with other groups. However, as the projects in this class require the use of
software tools and frameworks that students may have uneven prior familiarity with, dis cussion and assistance among students in gaining expertise with these software tools con stitutes acceptable behavior. Note that this assistance and discussion cannot include
the sharing of access to any code produced in solution to the project assignments. In
order to avoid potential ambiguity in what constitutes ”code produced in solution to the
project assignment,” students wishing to aid their peers with auxiliary supporting scripts,
mechanisms, or examples are directed to pass any such artifacts to the course staff for vet ting and possible inclusion on project-specific FAQs rather than share it with their peers
directly.
In this project, you are going to use SimpleScalar as the evaluation engine to perform
a design space exploration, using the provided framework code, over a 18-dimensional
processor pipeline and memory hierarchy design space (some of these dimensions are not
independent). You will use a 5-benchmark suite as the workload.
1. Project Goal
Your assignment is to, with an evaluation count limit of 1000 design points, explore the
design space in order to select the best performing design under a set of two different
optimization functions. These include:
1. The “best” performing overall design (in term of the geometric mean of normal ized execution time normalized across all benchmarks)
2. The most energy-efficient design (as measured by the lowest geometric mean of
normalized energy-delay product [units of energy delay product are joule-seconds]
across all benchmarks)
2. Background
2.1. SimpleScalar
SimpleScalar is an architectural simulator which enables a study of how different pro cessor and memory system parameters affect performance and energy efficiency. The
simulator accepts a set of system design parameters and an executable (workload) to run
on the described system. A wide range of system statistics are recorded by the simulator
as the executable runs on the simulated system. Once the framework in this project is
setup, interested readers can have a look at one of the log files in rawProjectOutputData
folder to view SimpleScalar output.
This project heavily uses SimpleScalar but most of the interface is abstracted out by a
simpler framework interface. Nevertheless, you can refer to this SimpleScalar guide for
details about parameters passed to SimpleScalar.
2.2. Design Space Exploration
Given a set of design parameters, Design Space Exploration (DSE) involves probing var ious design points to find the most suitable design to meet required goals. Follow this
quick reading about DSE before moving ahead.
DSE can be performed for different design goals. For example, one DSE may want to
find the best performing design whereas another DSE may be aimed at finding the most
energy efficient design. A more complex DSE may look for the best performing design
given a fixed energy budget.
An exhaustive DSE simply tries out all possible combinations of parameter values to
find the absolute best design. However, as the size of design space increases this approach
quickly becomes infeasible. Consider a 10-dimensional design space with 5 possible
values for each parameter and 2 minutes simulation time to evaluate a given design point;
an exhaustive search will take 5
10 ∗ 2min ≈ 37years.
A more intelligent DSE employs heuristics to intelligently prune down the design space
and to prioritize evaluation of more reasonable design points first. If the assumptions
employed by the heuristics are correct, the DSE will still result in the best design. On the
other hand. with a set of reasonably justified assumptions a heuristic can result in a “good
enough” design point.
2.3. Energy-Delay Product
Energy-Delay Product (EDP) is a metric which consolidates both performance and energy
efficiency.
EDP = total execution energy * execution time
Design A takes 100pJ to process an image in 100ms, EDP = 10000 units. Design B
takes 80pJ to process an image in 2000ms, EDP = 160000. Design A is clearly more
energy efficient, but it performs poorly as it incurs more execution time. EDP enables a
more holistic design comparison.
3. Our Heuristic
We define OurHeuristic as follows:
1. Design space dimensions can be labelled as either explored and unexplored.
2. Initially all dimensions are unexplored
3. Choose an unexplored dimension, exit if all dimensions are explored
3.1. Evaluate all possible design points by changing the value of this dimension
only
2
3.2. Fix value of this dimension by selecting the best design so far (consider
DSE goal)
3.3. Mark this dimension as explored
4. Go to step 3.
You should choose an unexplored dimension in step 3 based on your PSU ID Numbers
of students in the group, as follows.
DSE dimensions can be categorized in four major classes as follows:
1. Branch predictor (BP) configurations (i.e. branchsettings, ras, btb)
2. Cache configurations (i.e. {l1, ul2}block, {dl1, il1, ul2}sets, {dl1, il1, ul2}assoc)
3. Core configurations (i.e. width, scheduling)
4. Floating Point Unit (FPU) configuration (i.e. fpwidth)
Based on your ID numbers, you should calculate
( ID Number 1 + ID Number 2) mod 24
and then you should look at the Table 1 and start from the first category in the correspond ing row, and then second category, and so on.
For example, if your ID numbers are 9123456789 and 9111111111, the remainder
of its sum’s division by 24 is 12 and you should explore Core configs first, then BP
configs, then Cache configs, and then FPU configs at last.
Please note that the current implemented heuristic in generateNextConfigurationPro posal function is a simple heuristic as follows and you should extend it as explained
above. Current implementation starts from the leftmost dimension and explores all pos sible options for this dimension, and then goes to the next dimension until the rightmost
dimension.
4. Logistics
The set of possible points within the design space to be considered are constrained by the
provided shell script wrapper runprojectsuite.sh. All allowed configuration parameters
for each dimension of the design space are briefly described in the provided shell script.
runprojectsuite.sh shell script takes 18 integer arguments, one for each configuration
dimension, which expresses the index of the selected parameter for each dimension. All
reported results should be normalized against a baseline design with configuration param eters which already hard-coded in the framework.
Note that not all possible parameter settings represent a valid combination. One
of your tasks will be to write a configuration validation function based upon restrictions
described later in this document. Further, note that this design space is too large to effi-
ciently search in an exhaustive manner. Hence, a heuristic will be developed to specify an
order in which the design space will be explored.
The framework code will evaluate a fixed number of design points per run. This pa rameter cannot be changed. The key part of your task in this project is to implement
a heuristic search function that selects the next design point to consider, given either a
3
Table 1. Exploration Orders based on PSU ID.
( ID Number Sum) mod 24 1
st 2
nd 3
rd 4
th
0 BP Cache Core FPU
1 BP Cache FPU Core
2 BP Core Cache FPU
3 BP Core FPU Cache
4 BP FPU Cache Core
5 BP FPU Core Cache
6 Cache BP Core FPU
7 Cache BP FPU Core
8 Cache Core BP FPU
9 Cache Core FPU BP
10 Cache FPU BP Core
11 Cache FPU Core BP
12 Core BP Cache FPU
13 Core BP FPU Cache
14 Core Cache BP FPU
15 Core Cache FPU BP
16 Core FPU BP Cache
17 Core FPU Cache BP
18 FPU BP Cache Core
19 FPU BP Core Cache
20 FPU Cache BP Core
21 FPU Cache Core BP
22 FPU Core BP Cache
23 FPU Core Cache BP
performance, or an energy efficiency goal. Note that the framework code must be run
once for each of the optimization function options.
The framework, as given, provides functionality to enforce several, but by no means all,
of the validation constraints. It is your job to implement validation functions to enforce
constraints described throughout this document.
5. Framework
A sample run to use the provided framework can look something like this:
Extract project files archive and navigate to
project directory.
make clean
make
./DSE performance
Different components of the framework are invoked in the following order:
DSE (project binary) → runprojectsuite.sh (shell script) → SimpleScalar
4
DSE binary invokes runprojectsuite.sh script which in turn invokes SimpleScalar
simulator with appropriate arguments. Several log files are generated in project directory
on every invocation.
6. Anticipated Steps
These steps can serve as a high-level guideline to aid you during the project:
1. Enter ID in YOURCODEHERE.c.
2. Setup provided framework to get a set of results using the provided “unintelligent”
heuristic.
3. Implement validateConfiguration and generateCacheLatencyParams functions.
4. Implement OurHeuristic in generateNextConfigurationProposal for both opti mization goals (a well performing design and an energy efficient design)
5. Complete Report
7. Submission Requirements
1. Project report
2. Code implementations of missing or stub functions within the provided framework
7.1. Project Report
Your report must at conform to requirements listed in Appendix A. This report, data
contained within and their analysis will be the primary means of assessing this project.
7.2. Code Implementations
You will submit the source files (Makefile, runprojectsuite.sh, *.cpp and *.h) of your
implementation as a single tar archive for an audit of your implementation efforts. Ensure
that your code compiles on CSE machines without errors. You can make changes to
framework if you conclude that they are required. The following commands will be used
to compile and execute your code (followed by analysis of generated log files):
# Extract project files archive and navigate to
project directory.
make
./DSE performance
./DSE energy
Please note that running each of exploration modes could take more than one hour, so
please start as soon as possible so you could finish by the deadline.
8. Modeling Considerations
The Instruction Count (IC) for each benchmark is a constant. Thus, for performance, you
will be trying to optimize Instructions Per Cycle (IPC) and the Clock Cycle (CC) time.
Unless specified otherwise, the following modeling consideration have already been
implemented in the framework to calculate EDP. However, the provided information
may be used for explaining design space exploration results.
5
8.1. Clock Cycle Time
We will use the following very simplistic model for clock cycle time: The clock cycle
time is determined by the fetch width, number of floating-point units, and whether the
machine is in-order, or dynamic as follows:
• Dynamic, fetch width = 1: 115 ps + FPU delay
• In-order, fetch width = 1:100 ps + FPU delay
• Dynamic, fetch width = 2: 125 ps + FPU delay
• In-order, fetch width = 2: 120 ps + FPU delay
• Dynamic, fetch width = 4: 150 ps + FPU delay
• In-order, fetch width = 4: 140 ps + FPU delay
• Dynamic, fetch width = 8: 175 ps + FPU delay
• In-order, fetch width = 8: 165 ps + FPU delay
FPU delay is determined by the number of floating-point units as follows:
• count = 1: 5ps
• count = 2: 10ps
• count = 4: 20ps
• count = 8: 40ps
8.2. Power & Energy
8.2.1. Core Leakage Power
• Dynamic, fetch width = 1: 1.5 mW
• In-order, fetch width = 1: 1 mW
• Dynamic, fetch width = 2: 2 mW
• In-order, fetch width = 2: 1.5 mW
• Dynamic, fetch width = 4: 8 mW
• In-order, fetch width = 4: 7 mW
• Dynamic, fetch width = 8: 32 mW
• In-order, fetch width = 8: 30 mW
8.2.2. Floating Point Unit Power
• count = 1: 0.25 mW
• count = 2: 0.50 mW
• count = 4: 1 mW
• count = 8: 2 mW
8.2.3. Cache and Memory
Following list comprises tuples of format: [cache size or memory, access energy(pJ),
leakage/refresh power(mW)]
• 8KB: 20pJ, 0.125mW
• 16KB: 28pJ, 0.25mW
6
• 32KB: 40pJ, 0.5mW
• 64KB: 56pJ, 1mW
• 128KB: 80pJ, 2mW
• 256KB: 112pJ, 4mW
• 512KB: 160pJ, 8mW
• 1024KB: 224pJ, 16mW
• 2048KB: 360pJ, 32mW
• Main Memory: 2nJ, 512mW
8.2.4. Energy per Committed Instruction
• Dynamic, fetch width = 1: 10pJ
• In-order, fetch width = 1: 8pJ
• Dynamic, fetch width = 2: 12pJ
• In-order, fetch width = 2: 10pJ
• Dynamic, fetch width = 4: 18pJ
• In-order, fetch width = 4: 14pJ
• Dynamic, fetch width = 8: 27pJ
• In-order, fetch width = 8: 20pJ
8.3. Validation Constraints
You must implement these validation constraints in your code. Specifically, validate Configuration and generateCacheLatencyParams must be implemented properly.
1. The il1 (L1 instruction cache) block size must be at least the ifq (instruction fetch
queue) size (e.g., for the baseline machine the ifqsize is set to 1 word (8B) then
the il1 block size should be at least 8B). The dl1 (L1 data cache) should have the
same block size as your il1.
2. The ul2 (unified L2 cache) block size must be at least twice your il1 (and dl1)
block size with a maximum block size of 128B. Your ul2 must be at least twice as
large as il1+dl1 in order to be inclusive.
3. il1 size and dl1 size: Minimum = 2 KB; Maximum = 64 KB
4. ul2 size: Minimum = 32 KB; Maximum = 1 MB
5. The il1 sizes and il1 latencies are linked as follows (the same linkages hold for the
dl1 size and dl1 latency):
(a) il1 = 2 KB means il1lat = 1
(b) il1 = 4 KB means il1lat = 2
(c) il1 = 8 KB means il1lat = 3
(d) il1 = 16 KB means il1lat = 4
(e) il1 = 32 KB means il1lat = 5
(f) il1 = 64 KB means il1lat = 6
(g) The above are for direct mapped caches. For 2-way set associative add 1
additional cycle of latency to each of the above; for 4-way add 2 additional
cycles; for 8-way add 3 additional cycles.
6. The ul2 sizes and ul2 latencies are linked as follows:
(a) ul2 = 32 KB means ul2lat = 5
(b) ul2 = 64 KB means ul2lat = 6
7
(c) ul2 = 128 KB means ul2lat = 7
(d) ul2 = 256 KB means ul2lat = 8
(e) ul2 = 512 KB means ul2 lat = 9
(f) ul2 = 1024 KB (1 MB) means ul2lat = 10
(g) The above are for direct mapped caches. For 2-way set associative add 1
additional cycle of latency to each of the above; for 4-way add 2 additional
cycles; for 8-way add 3 additional cycles; for 16-way add 4 additional
cycles.
8.4. Miscellaneous Constraints
These constraints have already been specified in the framework. Have a look at Sim pleScalar invocation command in runprojectsuite.sh for an exhaustive list of specified
parameters. Moreover, any parameter not specified in runprojectsuite.sh will default to
SimpleScalar default settings.
• mplat is fixed at 3
• fetch:speed is fixed at 1
• ifqsize can be set to a maximum of 8 words (64B)
• decode:width and issue:width equal to your fetch:ifqsize
• mem:width is fixed at 8B (memory bus width)
• memport if fixed at 1
• mem:lat is fixed at 51 + 7 cycles for 8 word
8
Appendices
A. Project Report Minimum Requirements
Your report must at least answer the following prompts in the exact same order.
1. Describe in 100 words or less how the provided framework and its components
enable a design space exploration.
2. List the design point chosen by your DSE.
3. Fill out the following table as detailed below.
4. Plots as detailed below
5. Describe a more sophisticated heuristic which you expect will perform design
space exploration (limited by 1000 design points) more effectively to find a better
performing design (with respect to execution time).
6. Elaborate on any 2 new insights you gained while working on this project.
7. List of additional resources used (optional)
8. Additional information or comments (optional)
A.1. Table
In each cell specify the parameter value followed by why this value guided the DSE
closer to your optimization goal (for example: more ALUs allow extraction of more ILP
and increase performance). Make sure the parameters are in the exact order as they appear
in runprojectsuite.sh
Parameter Performance EDP
Param1 (i.e. width) Value = Value =
Why = Why =
Param2 (i.e. scheduling) Value = Value =
Why = Why =
... ...
ParamN Value = Value =
Why = Why =
A.2. Plots
The report should include the following four plots:
A. Line plot of normalized geomean execution time (y axis) for each considered de sign point vs. number of designs considered (x axis)
B. Line plot of normalized geomean of energy-delay product (y axis) vs number of
designs considered
C. Bar chart showing normalized per-benchmark execution time and geomean nor malized execution time for the best performing design
D. Bar chart showing per-benchmark normalized energy-delay product and geomean
normalized energy delay product for the most energy-efficient design found
These four plots must be labelled in your report corresponding exactly to num bering in the list above. Furthermore, axis in the plots should be properly labelled.
9
A.3. Other Guidelines
For clarity in the written report, when listing the best design points, please do not represent
them in terms of their index representations (e.g. 1 0 0 5 2 ...) and instead describe the
actual value used for each dimension in a table or similar presentation.
Points will also be assigned for following the guidelines and adhering to appropriate
levels of clarity, and style (and spelling, grammar, etc.) for a technical document.
10
B. Project FAQs
Q: What are the column headers for the .log file?
A: normalized EDP, normalized Execution time, absolute EDP, absolute Execution
time. The writes to both the .best and .log files are generated near the end of main.
Q: What are the column headers for the .best file?
A: Headers differ by line:
Line 1 headers: bestEDPconfig, normalized EDP of bestEDPconfig, normalized Exe cution time of bestEDPconfig, absolute EDP of bestEDPconfig, absolute Execution time
of bestEDPconfig, absolute EDP of Bench 0 on bestEDPconfig, normalized EDP of Bench
0 on bestEDPconfig, absolute EDP of Bench 1 on bestEDPconfig, normalized EDP of
Bench 1 on bestEDPconfig, absolute EDP of Bench 2 on bestEDPconfig, normalized
EDP of Bench 2 on bestEDPconfig, absolute EDP of Bench 3 on bestEDPconfig, nor malized EDP of Bench 3 on bestEDPconfig, absolute EDP of Bench 4 on bestEDPconfig,
normalized EDP of Bench 4 on bestEDPconfig
Line 2 headers: bestTimeconfig, normalized EDP of bestTimeconfig, normalized Ex ecution time of bestTimeconfig, absolute EDP of bestTimeconfig, absolute Execution
time of bestTimeconfig, absolute Time of Bench 0 on bestTimeconfig, normalized Time
of Bench 0 on bestTimeconfig, absolute Time of Bench 1 on bestTimeconfig, normal ized Time of Bench 1 on bestTimeconfig, absolute Time of Bench 2 on bestTimeconfig,
normalized Time of Bench 2 on bestTimeconfig, absolute Time of Bench 3 on bestTime config, normalized Time of Bench 3 on bestTimeconfig, absolute Time of Bench 4 on
bestTimeconfig, normalized Time of Bench 4 on bestTimeconfig
Q: Why are there only 18 configuration parameters when SimpleScalar (and the
project specification) list so many more?
A: There are 18 configuration variables, and more derived settings from those 18 con-
figuration variables, and still more settings that are fixed as constant (e.g. MPLAT). Given
the block size (set independently), associativity (set independently), and number of sets
(set independently), you can determine total cache size for the L1D and I caches and then
validate if the latency for that cache (set independently) is set correctly.
Q: What’s a quota error, why are half my output files empty, and why can’t I
make new files anymore?
A: It means you are out of disk space. Each run of this program produces a large num ber of intermediate output files for the evaluated design points. These are kept to speed
up subsequent evaluations of the same design point in future runs as a means of reducing
debugging/heuristic development time. Consider cleaning out your browser caches if you
are low on disk quota before performing a project run.
11