LABS 58
Objectives
Implement Full Search Based Variable Block Size Motion Estimation using MIPS ISA with the search pattern shown below.
Implement datapath components
Objective 1Logistics
The template file vbsme.s includes an overview of the algorithm and public test cases.
Submission deadline is on September 30th 2:00 PM.
Demonstration during lab hours will include running your program using public test cases. The private test cases will be run offline and feedback will be provided to you on D2L.
Submit only the vbsme.s on D2L.Include the overall percentage of effort of each team member in the file.
Objective 1 Rules and Grading Requirements
you must follow the zigzag access pattern
you are not allowed to use special registers in your implementation
s and t registers are the only registers to store values during the execution of the program
avoid the following commands: division, mod
add comments to code describing function of operations.
Objective 1 Scoring 250 points
13 public test cases test cases 0 and 14 are for debugging purpose, not to be graded 65 points
Private test cases 185 points
Objective 1 Assumptions
Frame size : 16×16 to 64×64, where x and y dimensions can be any integer between 16 and 64 can be square or rectangle
Window size: One of 7 specified dimensions in the vbsme.s.
Objective 1 Strategy
You should approach the problem with three tasks.
Task 1: implementation of the SAD subroutine that just computes the sum of absolute difference for a given window size.
Task 2: implementation of the address generation for reading the corresponding frame elements based on the current position in the frame.
Task 3: implementation of the zig zag patter scan diagonal, left, down moves.
I highly recommend starting with the C or your favorite language based implementation for each task. It is even better if you implement in three address operand form in C.
You should pass address of the first element for the window and the frame data.Than use window size as stride for accessing the subsequent rows of the frame data. Even though the test data seems to be a 2D array, to QtSpim it is a one dimensional array. After confirming the SAD for a given position is working properly, then write the routine that realizes the zig zag pattern.Debugging becomes manageable with a modular design.
Objective 1 Grading and Demo: When grading your vbsme implementation, we will use our vbsme.s template with different test cases. We will copy your code starting with the line that says insert your code here into the private test case template to run private test cases.
TA will come to your desk and then students will download the file from d2l.Submit your code by 2pm on September 30th to avoid late penalty.
Demo will be done only for public testcases.
All team members MUST be present during the demo for answering questions.
Demo should only be given to TAs.
During the lab session, TAs will give priority to demos.
No submission is allowed after the demo. If done, then it will be accounted as a late submission and redemo will be needed.
Objective 1 Penalty Conditions
Percent effort not reported 20 penalty
Late submission or late demonstration 10 per day
Submitting files in a folder or in compressed form ziptar.20 penalty
Changing the file name or extension. 20 penalty
Failing to demonstrate 80 penalty
Failing to add comments 20 penalty
Illegal register usage: max score 100pts,
Illegal scan pattern usage: maximum score 60pts
Each unattended lab without notice 20pts penalty.
Unable to answer questions about your implementation during demo 100 pts penalty
Common Questions on SAD Routine
Q1: How are the frame and window matrices stored? Are they stored in sequential memory slots?
Answer: n the sum of array declaration is :
N: .word 3N is the address which contains the loop count, 5
X: .word 2, 4, 7X is the address of the 1st element in the array to be added
SUM: .word 0SUM is the address that stores the final sum
in the memory data is stored sequentially 3 2 4 7 0
we use N, X , SUM to access the proper location.
Q2: Where to exit to? When we finish the final test, is there a specific location we are supposed to send the program? We are getting the attached error, which Im assuming is because a return register is being set to 0. What I am not sure of is if this is purposeful or if this is because our program is rewriting that register to 0 before it is used.
Answer: If you are using JAL instruction then make sure to use ra register to get back to proper retrun address. If you are not using the JAL then make sure to save return address on stack and use it to jump back. At the end of the program, you can either do as safe exit or go into an indefinite jump loop to same address. example:
loop: j loop
Q3. Will we ever have to worry about the size of the frame being greater than the size of the window?
Answer: The Frame should always be the same size as the window or larger.
Q4: In the word document, it says that we can assume the size of the frame will be between 16×16 and 64×64 where x and y can be any integer between 16 and 64. Test case 14 in the file vbsme.s has a frame size of 4×4. Will this test case be graded since it goes against the assignment description?
Answer:They test0 and 14 were given so that students can use them in initial stage for quick validation of codes and debug purposes.
Q5. How do we access an index of an array based on a variable value that we cannot hardcode for use in instructions like load word and store word?
Answer:
You need to calculate memory location corresponding to array index you want to read using base address of the frame or window, and their dimensions.
Q6. Calculating SAD values
Is it sufficient to calculate the sum of the entire array for both the window and the frame and then take the absolute value of the difference? Or is it better to calculate the sum of difference in each indices since two different framewindows can generate the same SAD value even if theyre not matching?
Answer: Taking sum first and then difference is not correct. You need to take difference first and then sum. Thats why it is called SAD: Sum of Absolute Difference.
Suppose you had the window:
1 0 0
0 0 0
0 0 0
And the Frame:
0 0 0
0 0 0
0 0 1
Taking the Sum of both Matrices, and then looking at their differences gives 0, while the SAD of this pair is actually 2.
Q7. Will all SAD values be 0?
I know the project says to find the minimum SAD value, but the test windows seem to have an exact match on the frames. Would it be better if we were to only look for the exact window?
Answer:
There may or may not be exact match.
Q8. Initializing Sum Abs Diff
Do we need to initialize the Sum of Abs Difference to a specific value? If you do not initialize it, then using a comparison for the minimum SAD will never execute because 0 will always be less than any computed SAD.
Answer:
You can initialize it to largest possible sad value or the first SAD value should be initialization.
Assuming the SAD value between the first window and frame is not the largest possible value, the first SAD calculation would become the minimum anyway.
Q9. vbsme, negative numbers? I noticed that none of the frames or windows given in the test cases have negative numbers. Is it safe to say we dont have to worry about getting negative numbers in any test cases when we demo our vbsme code?
Answer:
Correct no need to worry about negative values
Objective 2 Datapath Component Design and Simulation
Objective 2 Part 1:
Below is a list of datapath components needed in your project. Templates are in the DatapathComponents folder.
RegisterFile
DataMemory
SignExtension
Mux32Bit2To1
Follow the comments given in the source codes
Synthesize and conduct functional verification with postrouting simulation for each component
Exhaustively test your components
It is your responsibility to make sure that these data path components are working properly. We will not be testing the functionality of these components individually
No demonstration required for the datapath components
You should plan completing by Lab 9
Objective 2 Part 2:ALU Design Start this exercise after covering ALU Controller Topic in the class
Design and develop an ALU that supports all operations required by the given MIPS ISA.
Template for the ALU is in the DatapathComponents folder.
No demonstration required for ALU design.
Suggested ALU Design Approach
Study the given instruction list next page and identify all operations needed. Refer to the MIPS ISA Reference.
List all the arithmetic operations.
Draw the block diagram of the ALU module with inputs and outputs properly labeled with their bitwidth.
You need to revise the ALU outputs of the ALU32Bit.v from Part 1. You need to accommodate potential 64bit outputs from certain instructions, as well as deal with interpreting the outcomes of arithmetic operations needed by different types of branch instructions.Overflow detection is not needed.
Test your ALU in postrouting simulation for all the operations. There is no demonstration requirement for this task.
Table 1. Required MIPS Operations for the datapath design
Type
Instruction
Code
Type
Instruction
Code
Arithmetic
Add
add
Logical
And
and
Add Immediate Unsigned Word
addiu
And immediate
andi
Add Unsigned Word
addu
Or
or
Add Immediate
addi
Not or
nor
Subtract
sub
Exclusive or
xor
Multiply
mul
And Immediate
andi
Multiply Word
mult
Or immediate
ori
Multiply Unsigned Word
multu
Exclusive or Immediate
xori
Multiply and add word to Hi,Lo
madd
Signextend half word
seh
Multiply and subract word to Hi,Lo
msub
Shift left logical
sll
Data
Load word
lw
Shift right Logical
srl
Store word
sw
Shift Word Left Logical Variable
sllv
Store byte
sb
Shift Word Right Logical Variable SRLV
srlv
Load half
lh
Set on less than
slt
Load byte
lb
set on less than immediate
slti
Store half
sh
move conditional on not zero
movn
Move to Hi Register
mthi
move conditional on zero
movz
Move to Lo Register
mtlo
Rotate Word Right Variable
rotrv
Move from Hi Register
mfhi
Rotate word right
rotr
Move from Lo Register
mflo
Shift word right arithmetic
sra
Load Upper Immediate
lui
Shift Word Right Arithmetic Variable
srav
Branches
branch if greater than or equal to zero
bgez
SignExtend Byte
seb
branch on equal
beq
Set on Less Than Immediate Unsigned
sltiu
branch on not equal
bne
Set on Less Than Unsigned SLTU
sltu
branch on greater than zero
bgtz
branch on less than or equal to zero
blez
branch on less than zero
bltz
jump
j
jump register
jr
jump and link
jal
Common Questions
Q1. When it says list all of the arithmetic operations, does it mean that we should determine if an instruction requires add, multiply, subtract, etc. or does it mean list all of the operations we will need to implement the full set of instructions? The reason this task seems confusing to me is because we already know were going to have to handle add, multiply, subtract, and, or, and set on less than. I dont see what else we could really need in terms of arithmetic operations.
Im also confused about what the block diagram of the ALU should look like. Are we expected to have different controls for each one of the branch instructions to determine how to interpret the result of the ALU? This seems like it should be outside of the ALU, so are we supposed to draw more of the datapath?
Answer:
Make sure to follow these steps:
1 Table in the lab description lists all the instructions that needs to be implemented. Read it carefully.
2 Refer to MIPSInstruction reference manual to see how all of these instructions are actually executed and what ALU operations are needed to execute all these instructions. Make a list of these operations.
3 Narrow down on operations remove redundant operations. Identify all the unique operations.
4 Identify the width of the control signal you will need for your ALU to identify each of these unique operations.
5 Identify what all input and output signals you might need to add in your ALU if needed to implement these instructions.Adding new ports later in the project will need a lot of changes in your design, so its better to do it early in the stage.
6 Make a block diagram of your ALU on a sheet of paper with required inputs including control signal and outputs we are not looking for RTL designs. Note down all the operations that you will need to implement in ALU.
You need to identify all the operations you might need to include in your ALU for implementing all the instructions. For example, slt instructions will need to check for less than condition inside ALU and similar operator might be used by bltz depending on your implementation. Similarly, an Add operator can be shared among LW and ADD instructions.
Arithmetic instructions are different than operations that are discussed above. You have been asked to list all the unique operations you will need to implement in your ALU to include all the instructions in Table 1. Same operations can be shared between multiple instructions. READ MIPSInstruction manual and above responses carefully.
There are conditional branch instructions which will need ALU involvement. Further, you need to control zero flag inside your ALU to enable or disable branching. For Data instructions, you will need ALU to calculate the memory address you want to access in data memory. For instructions related to move you can use ALU to assign input to output.
Q2. In ALU32Bit.v , The outputs ALUResult and Zero were not declared as a reg type. I tried to assign these outputs values, and got an error procedural assignment to a non register is not permitted. I declared both outputs as a reg type and the error disappeared. My question is, whether or not we are meant to declare the outputs as reg type, and if not, how are we supposed to assign values to the outputs?
Answer:
Components files given to you is just a reference. You can change the data types and add or remove ports if you think you will need that in your implementation.
Q3.Im very confused for the instructions that involve the Hi and Lo registers. How do we access them when they are special registers and not in our register file component?
Answer:
Hi and Lo are two special registers outside the register file. each one is 32 bits wide.
when you multiply two 32 bit values you generate a 64 bit value.
higher order 32 bits and lower order 32 bits of the product are mapped onto the hi and loo registers respectively.
Q4. Can we read 4 registers from the regfile at once. Also can we write to to register in the regfile at once. ecause we need to read hilo and also read 1 and read 2 at the same time, so thats 4. And then we need to write into hilo, so 2 writes.The only other solution I could think of was putting hilo into their own register file
Answer:
HILO registers are not the part of the register file. They are internal registers for ALU explicitly designed for special instructions. You will be designing these registers in execute stage where as registerfile 32 addressable registers is a part of the decode stage.
Q5. MDM: Unconnected ports Address0 Address1 Address9 … Synthesis Warnings
Why does the input Address for the memory modules have to be 32 bits? In DataMemory, we only use bits 8:2 for addressing and in InstructionMemory, we only use 11:2 for addressing? I am worried that vivado will not handle these unused Address bits well and I am wondering why we even need them in the first place to be passed to the memory units in the first place. Why cant we send only Address8:2 into InstructionMemory and only Address11:2 into DataMemory? If I leave it as is, will these unconnected ports prevent the code from being successfully implemented on the FPGA?
Answer:
Vivado will trim of unused logic and ports during synthesis and implementation phase. For example, if you try to connect port131:0 to port211:0. Vivado will trim of port131:12 and will connectport111:0 to port211:0
Q6. Unroutable DSP cascade connection
I get the following error when trying to run implementation.
Place 30119 Unroutable DSP cascade connection found. Port PCOUT of DSP block ALUResult01 can drive only PCIN port of different DSP block.
Everything works perfectly in post synthesis simulation. Any idea why this is happening? I also have the warning from synthesis saying sequential elements are being removedall pipeline register outputs , but the simulation still works so Im not really sure what to do about that
Answer:
Check if you are using blocking assignment for the Madd and Msub operations and remove blocking assignments.
Q6.1: Currently we only have it blocking because we parse multTemp64 bits to HI and LO, so multTemp needs to update immediately for that to work. How could we calculate the multiplication first without blocking?
Answer: you need to add the previous value of HI and LO registers with current result of multiplication for Madd. Basically, you need to have an input port HILO which comes from the output of HI and Lo registers for ALU, and used that to complete Madd and Msub operations.
To avoid latches, make sure to give some default values to output and intermediate signals before entering into case or ifelse statements.
Q7. Inferring Latch unused sequential elements
what does inferring a latch mean? I have my case statement with an output for every case I have written, yet it says I am inferring a latch. I do not have every case possible written but every case I have written has an output. Also I am getting a warning about unused sequential elements, however the element it claims is unused is an element that I have connected back to a mux, it is the PC value that can get shifted but doesnt in this lab. Any ideas?
Synthesis Warning Ports drive by constant Zero
Answer: If all the modules are properly connected then it should not be an issue. Also make sure that all the instruction memory locations are initialized to default values. Initialize all the intermediate signals. Have you tried to open the Schematic? The wire could be showing that it is hooked up and also running to ground. In such cases you need to initialize the wire in the top level design.
Reviews
There are no reviews yet.