[SOLVED] COMP4601 Lab

Record your latency and utilization results for solutions 1 – 5 in two tables that allow these
to be compared. Do not cut and paste the synthesis reports for individual solutions into your
report.solution1
The matrix-vector multiplication code of Figure 4.4 is contained in
matrix_vector_base.c. This code can be loaded into the project
matrix_vector_proj using the Tcl script matrix_vector_proj.tcl and the Vivado
HLS command prompt: vivado_hls -f matrix_vector_proj.tcl
After you have created the project, run vivado_hls -p matrix_vector_proj from
the Vivado HLS command prompt to open the synthesized baseline solution of the
project in the Vivado HLS GUI.Describe the performance and resource utilization before adding any directives.
Briefly explain the execution schedule obtained.solution2
Create a new solution from solution1.
Unroll the dot_product_loop completely by adding a directive to the directive
script and run synthesis.
Compare the performance and utilization with solution1.
Explain why you obtain a loop iteration latency of 6.solution3
Create a new solution from solution1.
Pipeline the dot_product_loop with the default II and run synthesis.
Compare the performance and utilization with solution2.
Explain what happened to the loop nest.
Explain why the iteration latency is 3 and the iteration interval is 1.solution4
Create a new solution from solution1.
Unroll the data_loop and run synthesis.
Is this worth doing? Why (not)?solution5
Create a new solution from solution1.
Pipeline the data_loop and run synthesis.
Briefly explain the scheduling of the loop.
Taking both performance and utilization into account, rank the 5 solutions you have
so far in your order of preference and explain your choice.solution6
Create a new solution from solution1.
Compare the performance and utilization of the manually unrolled code of Figure 4.6
with that of solution2. (The code is in the file named matrix_vector_base_unroll_
inner.c. Copy this file to matrix_vector_base.c within your Windows directory and
reload the source file in the GUI to confirm that you have copied the code correctly.)
Run synthesis.
Record your latency and utilization results for solutions 7 – 11 in two tables that allow
these to be compared.solution7
Create a new solution from solution5.
In Windows, copy the file matrix_vector_base_copy.c to matrix_vector_base.c
so as to revert back to the code used for solution1-5. Reload the source file in the
GUI to confirm that you have restored the code correctly.Add array_partition directives to the M and V_In arrays while pipelining the
data_loop. The effect should be similar to the effect of, but not the same as, the
listing of Figure 4.11. Add the directives %HLS ARRAY_PARTITION variable=M
cyclic factor=2 dim=2 and %HLS ARRAY_PARTITION variable=V_In cyclic
factor=2 dim=1 to the directives script and run synthesis.
Compare the resulting performance and utilization with that of solution5.
Briefly explain the execution schedule.solution8
Create a new solution from solution7.
Modify the array_partition directives to use block partitioning and run synthesis.
Explain the observed performance in the light of solution7.solution9
Create a new solution from solution7.
Modify the array_partition directives to implement complete partitioning and
run synthesis.
Compare solution5, solution7 and solution9 in terms of performance and utilization.
Explain your findings.solution10
Create a new solution from solution9.
Modify the pipeline directive to target an II=2 and run synthesis.
Compare solution7, solution9 and solution10.solution11
Create a new solution from solution9 but set the target clock period to 5 ns. Run
synthesis.
Explain the loop iteration latency you observe.
Do you think there is any further improvement in performance possible?DFT exercises
Record your latency and utilization results for solutions 1 – 4 in two tables that allow these
to be compared.solution1
The DFT baseline code of Figure 4.15 is contained in dft.cpp1
. This code can be
loaded into the project dft_proj using the Tcl script dft_proj.tcl and the Vivado
HLS command prompt: vivado_hls -f dft_proj.tcl
After creating the project, run vivado_hls -p dft_proj from the Vivado HLS
command prompt to open the synthesized baseline solution of the project in the
Vivado HLS GUI.Describe the performance and resource utilization before adding any directives.
Briefly explain the execution schedule.solution2
Create a new solution from solution1.
Pipeline the inner loop labelled dft_label0 and run synthesis.
Explain the impact of pipelining dft_label0 on the performance, execution
schedule and utilization.Which operations limit the iteration interval?
1 There are some minor differences between the listing of Figure 4.15 and the contents of dft.cpp. IN_TYPE
and TEMP_TYPE were set to float, and the expression for w was altered to allow the use of ap_fixed type
data.
solution3
Create a new solution from solution2.
Change IN_TYPE and TEMP_TYPE to be of type ap_fixed<16,4> and run synthesis.
Describe the performance and utilization of the resulting design in comparison to
solution2.Outline the most significant constraints on the performance of this solution.
What considerations have you ignored in changing the program data types?
What could you do to assess the impact of changing the program data types?solution4
Create a new solution from solution3.
Swap the inner and outer loops of the source code as explained on pages 97-99. Run
synthesis.What do you observe? Why?
Include a copy of your loop interchange code into your report.