CMSC 216 Project #3

$30.00

Category: You will Instantly receive a download link for .zip solution file upon Payment

Description

5/5 - (8 votes)

1 Introduction and purpose
In this project you will write some functions to manipulate the instructions of a fictional simple processor named the MAD
Raisin. The MAD Raisin CPU has a limited set of instructions, is able to access just a small amount of memory, and is
designed for use in embedded systems, such as for example a toaster. (The computational needs of a toaster are not high.)
One purpose of the project is to use the features of C covered recently, namely bit operators and arrays, but another major
objective is to introduce hardware and assembly language concepts, which will be revisited later in the course.
Much of this project description explains the assembly and machine language concepts involved, as well as the hypothetical MAD Raisin CPU. The descriptions of the functions you have to write is just three pages of this assignment.
Although the amount of code to be written in this project is not large compared to some CMSC 132 projects you should
find it significantly harder than Projects #1 and #2, so start working on it right away. Before starting to code, look at
the public tests to see how the functions are called, and look at all of the definitions in the header file raisin.h described
below. Also first study bit masking from lecture, which you need to understand very well, and if needed, ask about it in
the TAs’ office hours before starting to code. Lastly, read the entire project completely and carefully before starting to
code. You may need to read it several times while writing the project.
Due to the size of the course it is not feasible for us to be able to provide project information or help via email/ELMS
messages, so we will be unable to answer such questions. However you are welcome to ask any questions verbally during
the TAs’ office hours (or during, before, or after discussion section or lecture if time permits). However, you cannot wait
to write projects in a course with ~600 students. If many students wait until the last few days to write or finish projects–
especially challenging ones– and have questions, the TAs will not be able to help everyone in office hours. If you are in
office hours near when a project is due you may have to wait a long time, and may not even be able to get help at all. This
would not be the TAs’ fault; it would be because you waited until too late to do the project, given the size of the course.
The only way to avoid this is starting projects early, in case you do have to ask questions in office hours.
1.1 Extra credit and number of submissions
If you make only one submission for this project that passes all the public tests we will give you 5 extra–credit bonus
points on the project. Additionally, if your single submission is made at least 48 hours before the on–time deadline you
will get 5 additional extra credit bonus points, for 10 extra credit points total. (Obviously you can’t get the second 5
extra credit points if you don’t get the first 5 extra credit points, but you can get the first 5 without getting the second 5.)
In order to get any extra credit you will have to read this assignment very carefully and throughly test your functions
yourself, so you are confident of doing well on the secret tests. If you find a bug later and submit again, you won’t get any
bonus points. You will also have to use good style all during your coding, because your single submission is of course the
one that is going to be graded for style. So read the course project style guide carefully.
However, as before, if you make too many submissions you may again lose credit. To avoid making submissions that
don’t compile you should compile your code and fix problems before submitting. And to avoid making submissions that
do not pass all the public tests you should run them yourself, check your output, and fix any problems, before submitting.
(If for some reason your code passes all the public tests on Grace, but doesn’t work when you submit it, so you have
to submit more than once– and this is not due to an error on your part or a bug in your code– you can talk with me
verbally in office hours about receiving the extra credit despite having more than one submission.)
2 Machine details and background concepts
This section explains many background concepts, while the following one describes the functions to be written.
2.1 Hardware concepts
For a program written in any computer language to be executed on a particular machine it must be translated to the
instructions that the machine’s processor can execute. These instructions are referred to as machine instructions, but we
usually just say instructions. Instructions manipulate information in memory and also in the processor’s registers. Memory
is used to hold data, such as variables, and programs themselves are also stored in memory in modern systems. Registers
© 2023 L. Herman; all rights reserved 1
are locations that store values like memory except they are internal to the CPU, so they are accessed much more quickly.
Registers temporarily store values that instructions operate upon, as well as the results of instructions.
The fundamental operation of any CPU is to get an instruction from memory, figure out what it says to do, perform
that operation, and go on to get the next instruction. This is called the fetch–decode–execute cycle. (This is discussed in
more detail in the Bryant & O’Hallaron reserve text.) Although the instruction sets of different processors vary widely,
instructions can be categorized by functionality, for example:
computation instructions: These perform various arithmetic and logical operations.
flow of control instructions: By default instructions are executed sequentially in memory, but there are instructions that
affect which instruction is executed next, for implementing conditionals and repetition, as well as function calls.
data movement instructions: These transfer data values between memory and registers, between different registers, and
sometimes between different memory locations.
invoking the operating system: These instructions are used to do things like halt a program or to access parts of the
hardware that user programs are not allowed to directly access, for example, I/O devices. Some instructions allow
user programs to make system calls to get the operating system to perform these types of tasks for them.
2.2 Machine specifications
The hypothetical MAD Raisin processor has a 32–bit (4 byte) word length, meaning that instructions, registers, and memory
words are all 32 bits. (As mentioned in lecture, processors manipulate information in multibyte quantities called words.)
Machines that use the MAD Raisin CPU are quite small, with very limited system resources. They have only 65536 bytes
of memory, grouped as 16384 four–byte words (16384× 4 = 65536). The first byte in memory has address 0. Memory
addresses always refer to bytes, but memory is only word–addressable, meaning that although each byte in memory has its
own unique address, the CPU can only access memory in four–byte quantities, using the address of the first (or low–order)
byte of a four–byte word. Consequently, the addresses that can be used to access memory are 0, 4, 8, etc., up to 65532.
The value of a memory word can be interpreted as an instruction and executed, or treated as data.
The MAD Raisin CPU has 22 different instructions. It has eighteen registers, each of which as mentioned holds 32
bits. The registers are numbered 0 through 17, and are referred to using the names R0 through R17 (note there is an enum
with these values in raisin.h). The MAD Raisin has what is called a load/store architecture, which means that only some
data movement instructions access memory. One such instruction loads a value from a memory location into a register
and another one stores a value from a register into a memory location, while all other instructions, including computation
instructions, operate upon values in registers and place their result into a register. So computation involves loading operands
from memory into registers, doing calculations with them, and storing the results from registers back into memory.
Two of the MAD Raisin’s registers have special purposes. Register R16 always contains the value zero (all bits zero).
Because this is such a common value needed in computations it saves time for the CPU to always keep it in a register.
Register R17 is the program counter, which always contains the memory address of the next instruction to be fetched and
executed. A flow of control instruction can modify the program counter to cause execution to jump to somewhere else in a
program. All other instructions cause the program counter to be incremented by 4. So unless specified by a flow of control
instruction, the processor executes instructions sequentially in memory. The value of the program counter register can be
used by instructions but it cannot be directly modified by storing the result of a computation in it. And instructions can also
not modify the value of the zero register R16. The other registers may be used or modified by the various instructions.
2.3 Hardware, components of instructions, and instruction format
Since the MAD Raisin’s processor has a 32–bit word length, all instructions, just like all data, must fit into 32 bits.
When a 32–bit word is interpreted as an instruction it is considered to consist of fields representing (a) an “opcode”, i.e.,
operation of the instruction, (b) up to two registers used as operands by some instructions, and (c) a memory address or
constant/immediate field that holds either a memory address or a constant (literal) numeric operand for some instructions.
The instruction format on the MAD Raisin processor is as follows:
5 bits 5 bits 5 bits 17 bits
opcode register1 register2 memory address or constant/literal value
The fields have the following uses and functionality:
© 2023 L. Herman; all rights reserved 2
opcode: An opcode uniquely identifies the operation that an instruction performs. Short names are used in assembly
language to refer to each type of instruction. (Recall that assembly instructions are just human–readable versions of
machine instructions.) Since the MAD Raisin has 22 different instructions (opcodes), 5 bits (the leftmost 5 bits if the
instruction word) can represent any of them. But since 5 bits can store 32 different values and the MAD Raisin only
has 22 instructions, there are ten five–bit values that the opcode field can contain that do not represent valid opcodes.
register1, register2: As Section 2.4 says, some instructions operate upon one or two register operands. These two fields
of instructions contain numbers indicating which registers instructions use as operands. Since the MAD Raisin CPU
has 18 registers, 5 bits can indicate any of them. Since 5 bits can store 32 different values, but the MAD Raisin only
has 18 registers, there are fourteen values that can be in a register field that are not valid register numbers.
address or immediate (constant): Some of the MAD Raisin’s instructions have the address of a memory location as
an operand. For example, an instruction may copy the value in memory location 216 to one of the registers, so 216
would be contained in this field of the instruction. One instruction (named li) has a constant or literal value, typically
called an immediate in assembly language, as an operand. For example, an instruction may store the numeric value
250 in one of the registers, so 250 itself would be encoded in the instruction.
Instructions that have a memory address operand (see the table in Section 2.4 below) only use the rightmost 16 bits
of this 17–bit field. (16 bits suffice since the MAD Raisin has 65536 bytes of memory, and 216 = 65536). But the li
instruction, which contains an immediate value, uses all 17 bits of this field for an immediate, so immediate values
can be between 0 and 217 −1 (= 131,071). (Although literals that are parts of instruction words are limited in size
to this range, a larger value can be computed in a register or stored in a memory location, since a 32–bit word or
register can store values up to 4,294,967,295.)
For example, say an instruction has opcode value 3 and is an instruction that uses both register operands, and this
particular instruction is going to operate upon registers R3 and R5. Also suppose this instruction doesn’t use the address/constant field, and just has zeros in those 17 bits. Then this instruction would be represented in a 32–bit memory
word as 0x18ca000016 , which is 41589145610. (Suggestion– convert the hexadecimal value to binary and match the bits
against the diagram of the instruction format above.) Or suppose an instruction has opcode value 19 and uses one register
operand, which is R7, and uses the memory address field to store 216010. If it also had 0s in the unused fields it would be
represented in a memory word as 0x99c0087016 (which is 257949912010 ).
2.4 Names, operands, and effects of the machine instructions
A basic understanding of the effects of the MAD Raisin’s machine instructions is needed to write the project, so after the table below, which indicates which operands are used by each instruction, their effects are briefly described. Each instruction
is listed with the (decimal) number between 0 and 21 that represents its opcode, according to the enum opcode_name defined
in raisin.h. Note that the column headings register1 and register2 are the two register operand fields, and “addr/imm”
refers to the memory address/immediate operand field. A checkmark means that an instruction uses that field.
opcode value register1 register2 addr/imm
halt 0
syscall 1
add 2 X X
sub 3 X X
mul 4 X X
quot 5 X X
mod 6 X X
and 7 X X
or 8 X X
neg 9 X
not 10 X
opcode value register1 register2 addr/imm
eql 11 X X
neq 12 X X
lt 13 X X
le 14 X X
gt 15 X X
ge 16 X X
move 17 X X
lw 18 X X
sw 19 X X
li 20 X X
branch 21 X
Different instructions use between 5 bits and 27 bits of their word (there are no instructions that use all 32 bits). Unused
fields may just contain any bits at all. Also note that not every 32–bit word represents a valid MAD Raisin instruction.
Here are brief explanation of the effects of the instructions:
© 2023 L. Herman; all rights reserved 3
halt: This instruction is one of two that has no operands. It simply stops executing a MAD Raisin program and returns
control to the operating system.
syscall: This instruction, which also has no operands, allows a program to invoke a “privileged” function in the operating
system, such as performing input or output. (This is not explained further here.)
add: This instruction performs addition. (How the computation instructions use their operands is explained below this
list.)
sub: This instruction performs subtraction.
mul: This instruction performs multiplication.
quot: This instruction performs division. (“quot” stands for “quotient”.)
mod: This instruction computes the remainder of a division (modulus).
and: This instruction performs logical conjunction.
or: This instruction performs logical disjunction.
neg: This instruction performs arithmetic negation, changing a positive value to a negative one or vice versa.
not: This instruction performs logical negation, changing a true value to a false one or vice versa.
eql: This instruction compares the values in its two register operands for equality.
neq: This instruction compares the values in its two register operands for inequality.
lt: This instruction compares the values in its two register operands for less than.
le: This instuction compares the values in its two register operands for less than or equal.
gt: This instruction compares the values in its two register operands for greater than.
ge: This instruction compares the values in its two register operands for greater than or equal.
eql, neq, lt, gt, le, and ge, in conjunction with the branch instruction explained below, allow execution to go to
another location in a program. If the relationship they are testing is true about the values in their register operands
they will set a “condition code flag” in the CPU, and clear the flag is the relationship is not true. The status of the
condition code flag can be used by the branch instruction as explained below.
move: This instruction copies its second register operand’s value to its first register operand (without modifying the value
in the second register operand).
lw: The lw instruction (load word) copies a value from a memory location to a register. It uses the first register operand
and a memory address, and its effect is to copy the value of the word in memory that has that address into the register
specified by the register operand. For example, lw R3 2000 would load or copy the value that is in memory location
2000 into register R3. (Assuming 2000 is in decimal, this instruction would be stored in a register or memory word
as 0x90c007d016 , which is 242850401610 .)
sw: The sw instruction (store word) does the opposite of lw, copying a value from a register to the indicated memory
location.
lw and sw are the only instructions that access or modify values in memory locations, because the MAD Raisin has
a load/store architecture.
li: This instruction loads a constant (or literal, or immediate) value (li stands for load immediate) into a register. For
instance, li R4 2368 would load the numeric value 2368 (not the contents of memory location 2368) into register
number R4. (Assuming 2368 is in decimal, this instruction would be stored in a word or register as 0xa100094016 ,
which is 270113414410 .)
branch: This instruction examines the value of the CPU’s internal condition code flag, and treats the 17–bit value in
the memory address or immediate field as an address. If the condition code flag is set then this instruction causes
the value in the memory address/immediate field to be copied into the program counter register (R17), causing
execution to jump to that address next, and continue from there. If the condition code flag is not set then the program
counter register is not modified, and execution just drops down instead to continue with whatever instruction follows
sequentially after the branch instruction. As mentioned above the condition code flag is set by the six comparison
instructions eql, neq, etc.
© 2023 L. Herman; all rights reserved 4
All the instructions are computation instructions except move, lw, sw, and li, which are data movement instructions;
branch, which is a flow of control instruction, and halt and syscall, which invoke the operating system.
The instructions add, sub, mul, quot, mod, and, and or perform the operation described above to the values in their two
register operands and put the result in their first register operand. For example, the add instruction sums the contents of
register2 and register1 and puts the result in register1. The instruction div R4 R3 divides the value stored in register R4 by
the value stored in register R3 and puts the result in register R4. And sub R10 R12 subtracts the contents of R12 from the
contents of R10 and puts the result into R10. In fact, all instructions that modify a register store their result into their first
register operand. Since a register can only store one value at a time, any instruction that modifies its first register operand’s
value has the effect of overwriting the previous value in that register.
The instructions eql, neq, lt, le, gt, and ge perform the operation described above to the values in their two register
operands but just set the CPU’s internal condition code flag, without modifying their first register operand.
Note that the same register may be used for both operands of an instruction that uses two registers, e.g., add R3 R3.
The two computation instructions that have one register operand, neg and not, perform the operation described above
to the value in their only register operand and put the result in the same register operand. For example, neg R7 would
perform arithmetic negation of the value in register R7 and store the result back in R7.
Any operand fields that aren’t indicated in the table above are ignored by that instruction. For example, move has two
register operands, therefore they will be contained in the two register fields register1 and register2; the address/immediate
field is not used by a move instruction. And a halt or syscall instruction doesn’t use any operand fields.
Of the instructions that use the address or immediate field, only li uses that field to store an immediate (i.e., constant)
operand; the others use that field to store a memory address.
Note the MAD Raisin is such a simple machine it can only manipulate integer data, not characters, floats, etc.
3 Functions to be written
The header file raisin.h contains definitions and the prototypes of the functions you must write. The functions use an
unsigned int to represent a MAD Raisin word. On the Grace machines an int is four bytes.
3.1 unsigned short print_instruction(unsigned int instruction)
This function should try to interpret its parameter as a MAD Raisin instruction, use C’s bit operators to extract its fields
(see the description and diagram in Section 2.4), and print the instruction’s components on an output line. However, if its
parameter doesn’t represent a valid MAD Raisin instruction, according to the criteria discussed in Section 4 below, nothing
at all should be printed and the function should just return 0. Otherwise, if the instruction is valid, the function should
return 1 after printing the instruction in this format:
• The instruction’s opcode (the opcode field of the parameter, meaning its leftmost five bits) should be printed using
its name in the table in Section 2.4 (halt, syscall, add, sub, etc.), spelled exactly as shown there. For example,
if the value of the opcode field is the enum value HALT (which has the value 0), then halt should be printed. If the
opcode field is ADD then add should be printed, etc.
• Following the opcode, the register operands that are actually used by the instruction should be printed, in the
order register1 followed by register2 if an instruction uses both register operands. For example, an and instruction
uses both registers as operands, so both should be printed, with the first register operand followed by the second one,
but a neg instruction uses the first register as its only operand, so just the first register operand should be printed.
Register names should be printed in decimal with an R immediately preceding the register number. For example,
register values 0, 1, and 12 would be printed as R0, R1, and R12.
• If an instruction uses a memory address operand or an immediate operand that operand should be printed last,
in decimal. If the instruction uses the memory address/immediate field for storing a memory address its value
should be printed using exactly five places, with addresses less than 1000010 printed using as many leading zeros
as necessary so they occupy exactly five places. For example, the address 21610 should be printed as 00216. (This
can be done the hard way, but note that it is trivial to accomplish in C with printf() formatting options covered
in discussion.) But for the li instruction, which uses the memory address/immediate field for storing an immediate
(constant) value, the field’s value should not be printed with any leading zeros (except zero itself should be printed
as 0). Only memory address operands should have leading zeros if needed.
© 2023 L. Herman; all rights reserved 5
The printed fields must be separated with one or more tabs or spaces; the exact number of tabs or spaces (or both)
doesn’t matter and is up to you. The printed instruction should not have any tabs or spaces before the opcode or following
the last field printed. A newline must be printed after the instruction.
For example, the call print_instruction(0x6a4e0000) should print something like lt R9 R7, where the exact
number of spaces or tabs separating the three things printed is up to you, as long as it is one or more.
(Remember that there is no such thing as a hexadecimal, decimal, or octal number in memory. There are just numbers
in binary, although programmers can use constants in different bases in programs, and programs can read or write numbers
in different bases. So even though the argument passed into this function above is in hexadecimal, it is stored in memory
in binary, and you can perform bit operations on it to extract its fields.)
Note: in projects where output has to be produced, students sometimes lose credit for minor spelling or formatting
mistakes. The submit server checks your output automatically, consequently even trivial typos would cause tests to fail.
Due to the size of the course we would not be able to give back any credit lost for failed tests due to spelling or formatting
errors, even if they are extremely minor, because that would necessitate manual checking of every output for ~600 students’
projects, which is not feasible. Therefore it’s up to you to check carefully that your output is correct, register names and
opcodes are spelled exactly, and that the output format described above is followed.
3.2 int load_program(unsigned int memory[], const unsigned int program[],
unsigned int start_addr, unsigned short pgm_size,
unsigned short data_segment_size)
This function simulates the operating system loading a machine language program into the MAD Raisin’s memory so it
could be executed. It should copy the contents of the array program (representing the machine language program) into
the array memory, beginning at the address start_addr. (Note that start_addr is a MAD Raisin byte address, not an
array subscript or a number of array elements. For example, if the program should be loaded starting at the fourth word of
the MAD Raisin’s memory the value of start_addr will be 12, because each word is four bytes.) The number of words
in program is given by pgm_size. However, not all of the words of the program might be instructions. Each program
has an area of its memory for its actual code (called the text segment) and another area for its data or variables (called
the data segment). (Note that although they are much more complex this is the case for real machines, for example, see
slide #15 in Lecture #1, where the data segment consists of both regions labeled “Dynamic data (heap)” and “Global and
static data”.) The parameter data_segment_size represents the size of the program’s data segment, which will always
be at the end of its memory. For example, if pgm_size is 100, meaning the program has 100 total words of memory, and
data_segment_size is 25, this means that the program’s first 75 words are instructions and its last 25 words are data.
If its parameters are valid the function should load the program into memory starting at address start_addr and return
1. If its parameters are invalid it should not modify anything and just return 0. Its parameters would be invalid if:
• start_addr is larger than the maximum word address on the MAD Raisin. The value of the symbolic constant
NUM_WORDS defined in raisin.h gives the number of words on the MAD Raisin, so the maximum word address can
be determined based on that (remember the first word has address 0).
• start_addr is not a valid word address that is divisible by 4.
• The size of the program in program, given by pgm_size, is larger than the number of words in the MAD Raisin’s
memory.
• Copying pgm_size words starting at the address start_addr would go beyond the end of the MAD Raisin’s memory.
(Note that it’s not invalid if pgm_size is zero; in that case no instructions would be copied to the memory array, and
the function would just return 1.)
• The values of the parameters pgm_size and data_segment_size would mean that there is not at least one instruction
in the program’s text segment. (A nonempty program cannot have no instructions, and it cannot consist only of data–
there has to be at least one instruction.)
The function should not modify the array memory outside of storing the indicated number of words into it beginning at
the indicated address. When a program is loaded there may be values in other memory locations left from prior programs;
those just remain as garbage values. Note that nothing prevents one program from being loaded into memory locations
where a different program had previously been loaded.
Since its parameter memory represents the memory of the MAD Raisin, the function may assume that it has NUM_WORDS
elements. A C function has no way to test whether an array passed in is the right size; this is up to the caller to ensure. The
function may also assume that the array program has at least pgm_size elements (which it also cannot check itself).
© 2023 L. Herman; all rights reserved 6
3.3 unsigned short disassemble(const unsigned int memory[], unsigned int start_addr,
unsigned int pgm_size, unsigned int data_segment_size)
A disassembler converts machine language to assembly language, which is the reverse of what an assembler does. This
function will do that for a MAD Raisin program stored somewhere its first parameter memory, which is an array of
words representing the MAD Raisin’s memory. The function may assume that memory is a valid C array that has at
least NUM_WORDS elements. The program to be disassembled begins at the MAD Raisin address start_addr in the array. (As in Section 3.2 above, start_addr is a byte address, not an array subscript.) The parameters pgm_size and
data_segment_size have the same meanings as they do in the description of load_program() in Section 3.2 above.
If its parameters are valid the function should print the program in the format described below and return 1. If its
parameters are invalid it should not produce any output at all and just return 0. What would make the parameters invalid
is described below. (Look at some of the public test outputs for examples.)
• Each word in the array memory that represents an instruction should be printed exactly the same way that the function
print_instruction() (Section 3.1) prints instructions, with each one followed by a newline.
• Then the function should print any subsequent elements of the array that represent data, based on the value of
data_segment_size, in hexadecimal. Values of memory words that are less than 1000000016 should be printed
using enough leading zeros so they occupy exactly eight places. The values should be printed in the way that
printf() prints hexadecimal values using the %x format specifier. (A leading 0x should not be printed.)
Instructions must be valid but there are no validity checks for elements in the data segment, because data can have any
value. Note that if its parameters are valid the function will produce exactly pgm_size lines of output, but not all of them
will necessarily be printed instructions– if data_segment_size is greater than 0 the last data_segment_size lines will
represent data and just be printed in hexadecimal.
The function’s parameters would be invalid if:
• Any of the words of memory, starting at the address start_addr (recall that it is a byte address, not an array
index), which represent instructions (not data) are invalid, using the criteria described in Section 4 below.
• The size of the program in program to be printed, as given by pgm_size, is larger than the number of words in the
MAD Raisin’s memory. (It’s not invalid if pgm_size is zero; in that case nothing would be printed, and the function
would just return 1.)
• If printing pgm_size words starting at the address start_addr would go past the end of the MAD Raisin’s memory.
• If the program is nonempty (based on the value of pgm_size) and the values of pgm_size and data_segment_size
would mean that there is not at least one instruction in the program’s text segment. (A nonempty program must have
at least one instruction, but it’s not invalid if a program is completely empty, or if the data segment has no elements.)
4 Instruction validity
As mentioned, some values that can be stored in a MAD Raisin word or register don’t represent valid machine instructions.
The reasons that a word could be invalid as a machine instruction are:
• The value in its opcode field (leftmost five bits) is wrong. The MAD Raisin has 22 instructions (hence opcodes) on
(with enum values in raisin.h between HALT and BRANCH), so values outside that range don’t represent opcodes.
• If a register operand that is actually used by the instruction has an invalid number. There are only eighteen registers
on the MAD Raisin with numbers 0–17 (and enum values R0 through R17 defined in raisin.h), so values that can
fit into the five–bit register fields that are outside that range don’t represent valid register numbers.
• If it is an instruction that uses its address/immediate field (rightmost 17 bits) to store a memory address and the
value in the field is not evenly divisible by 4. Since everything on the MAD Raisin occupies a 4–byte word, and all
data and instructions are accessed only in whole–word units, the only valid memory addresses are divisible by 4.
Note: if the parameters represent an instruction that uses the address or constant field for storing a memory address
its value must be divisible by 4. But if the instruction is an li instruction, which is using the field to represent an
immediate (constant) value instead, it does not have to be divisible by 4.
• If the parameter represents an instruction that modifies its first register operand’s value and the instruction’s register1
field contains the number of one of the special unmodifiable registers (the program counter register R17 or the zero
register R16). The instructions that use and modify the first register operand are indicated in Section 2.4.
© 2023 L. Herman; all rights reserved 7
R16 and R17 could be used as the second register operand of any instructions that use two registers, or as the first
operand of instructions that use but don’t modify their first register operand, but not as the first operand of instructions
that would modify their first register operand’s value.
Note: the values of fields that are not used by instructions have no effect on validity, and it’s immaterial what they
contain. For instance, a neg instruction may have anything at all in its rightmost 22 bits (the second register operand field
and the memory address/immediate field). As long as the values in its opcode and register1 fields are OK, a neg instruction
is valid. But instructions that do use fields, and have incorrect values in them, are invalid.
A Development procedure review
A.1 Obtaining the project files, compiling, checking your results, and submitting
Log into the Grace machines and use commands similar to those from before:
cd ~/216
tar -zxvf ~/216public/project03/project03.tgz
This will create a directory named project03 that contains the necessary files for the project, including the header file
raisin.h and the public tests. You must have your coursework in your course disk space for this class (the cd command
above), otherwise your submission will not be accepted. After extracting the files from the tarfile, cd to the project03
directory, create a file named raisin.c (spelled exactly that way) that will #include the header file raisin.h, and in it
write the functions whose prototypes are in raisin.h.
As in Project #1, but unlike Project #2, your code will not be a standalone executable; you are writing functions that will
be compiled with and called from our tests, which contain main() functions. Consequently each public test is a different C
source file and will be compiled to form a different executable program. You can compile the tests like you did in Project
#1, either from the command line or under Emacs. For example, to compile your code with the first public test use:
gcc public01.c raisin.c -o public01.x
(Or you could use separate compilation and link the resulting object files together if you want.) Adjust the command
to use public02.c instead to compile your code for the second public test, etc.
Commands similar to those in Project #2 can be used to run your code and determine whether it passes a public test,
except there are no input files in this project, because the tests do not read any data. For example:
public01.x | diff -b – public01.output
If no differences exist between your output and the correct output then diff itself will not produce any output, meaning
that your code passed the test, but the -b argument to diff causes it to ignore differences only in the amount of whitespace,
because it’s up to you how much whitespace print_instruction() prints between components of instructions. (Of course
to see your output just run without the diff command, as just public01.x.)
As before, running submit from the project directory will submit your project. Before you submit you must make sure
you have passed the public tests, by compiling and running them yourself, using commands like the ones above.
A.2 Grading criteria
Your grade for this project will be based on:
public tests 40 points
secret tests 45 points
programming style 15 points
Almost half of your score will come from secret tests. The public tests check only a small subset of the functionality
of your code. If you don’t write extensive tests yourself you could lose substantial credit on the secret tests.
Style will still be a significant part of your score. If you want to avoid losing credit for style in this project and the
remaining ones, carefully read the course project style guide handout. It describes the style criteria in detail. The only file
that will be graded for style is raisin.c.
© 2023 L. Herman; all rights reserved 8
B Project–specific requirements, suggestions, and other notes
• Write your own tests first, before even trying to run your code with the public tests. Start with small tests that
each test just one thing. As you ensure your code doesn’t have bugs, or as you fix bugs, add more tests that test more
things.
• You may want to write some or all of the functions first assuming that all instructions are valid, then, when you have
things working well that way, go back and figure out how to check instructions for validity where needed.
Alternately, you may first want to figure out how to write code to check instructions for validity and get this code
to work, before even writing any of the functions, then call it when the required functions have to check instruction
validity.
• Unless you have versions of all required functions that will at least compile, your program will fail to compile at all
on the submit server. (Suggestion– create skeleton versions of all functions when starting to code, which just have
an appropriate return statement, like we did in the skeleton functions.c in Project #1.)
• You cannot modify anything in the header file raisin.h or add anything to raisin.h, because your submission
will be compiled on the submit server using our version of it.
Your code may not comprise any source (.c) files other than raisin.c, so all your code must be in that file. You
cannot write any new header files of your own either.
Do not write any code in raisin.h– your code must be in the file raisin.c. Header files are for definitions, not
executable code. Your code will not compile on the submit server unless all of your code is in raisin.c.
Do not write a main() function in raisin.c, because your code won’t compile in that case (since our tests already
have main() functions). Write any tests in your own separate source files, and compile them together with raisin.c.
(Don’t modify the public tests, otherwise you won’t be testing your code on the cases that the submit server is
running. Just create your own tests in different source files.)
• As the project policies handout on ELMS says you should use only the features of C that have been covered so far,
up through when this project is assigned. (As in Project #2, this includes C operators in Chapter 5 of the Reek text,
and any C features in Chapter 4 of the textbook, except you cannot use the goto statement.) Note you may not use
bit fields of structures (Section 10.5 in the Reek text); they will not be covered at all this semester, and that chapter
has not been covered yet anyway.
• One purpose of the project is to use and get experience with C’s bit operators and to understand bit masking. To
avoid losing credit you must use the bit operators anywhere you have to access or manipulate parts of words. Do
not multiply numbers, or divide them, or use exponentiation, to manipulate the bits of numbers– use the bit
operators instead. Similarly, when you need to extract or manipulate a part of a number do not shift it in both
directions to isolate only some of its bits– use bit masking instead. If you have questions about how to perform
bit operations on words or to extract parts of words ask in the TAs’ office hours. (Read this item again.)
• Note that the project style says that global variables may not be used in projects unless you are specifically told to
use them. You will lose credit for using any global variables in this project. (Read this item again also.)
• If your code compiles on Grace but not on the submit server either you modified raisin.h or your account setup
is probably wrong. To check the way your account is set up just run check-account-setup, and come to the TAs’
office hours for help if you can’t fix any problems that it identifies yourself.
If your code passes tests on Grace but not on the submit server, or if it passes tests sometimes but not other times,
remember that the third lecture of the semester emphasized that uninitialized variables can cause C programs to work
differently on different machines or different times they’re compiled, and that uninitialized data is the most common
source of bugs in this course. Your TA recently showed a program check-vars that can say whether your program
is using any variables that have not been previously initialized, which you can use to see if this is the case.
• When you are writing your own tests you will need to create MAD Raisin instructions. Suggestion: do not use
decimal to write instructions. Write the binary representation of an instruction according to the instruction format
in Section 2.3, then convert it to either octal or hex. It is straightforward to convert binary to octal or hex, but it is
inconvenient to convert large binary numbers to decimal.
© 2023 L. Herman; all rights reserved 9
It is good practice (and not difficult) to convert manually from binary to hexadecimal. But note that if you use a
conversion program to convert between bases, some of them will drop leading zeros, which can be confusing.
• For this project you will lose one point from your final project score for every submission that you make in excess
of five submissions. You will also lose one point for every submission that does not compile, in excess of three
noncompiling submissions. Therefore be sure to compile, run, and test your project’s results before submitting it.
Hopefully everyone will check their code themselves carefully, and avoid these penalties.
• Keep in mind that that the project policies handout says that if you submit more than once your last submission
may not be the one that is graded, so use good style (according to the style guide) from the beginning when you
start working on the project, and throughout all of your coding.
• In just a few keystrokes Emacs can reindent an entire program, as your TA probably showed recently (and as is
explained in the UNIX tutorial). Using this simple Emacs reindenting command would allow you to be certain to
avoid losing any credit for that aspect of style grading for your project.
• Make sure none of your program lines have length more than 80 by running your Project #2 line length check
programs on raisin.c! If your project02 and project03 directories are both in the same parent directory and you
are located in the project03 directory, the command ../project02/lengthwarning.x < raisin.c would work.
The Project #2 secret tests will be put on Grace soon so you can fix any bugs in your lengthwarning program and
ensure its results are accurate.
• Recall that the course project policies handout on ELMS says that all your projects must work on at least half of the
public tests (by the end of the semester) for you to be eligible to pass the course.
• If you have a problem with your code and have to come to the TAs’ office hours, you must come with tests you have
written yourself– not the public tests– that illustrate the problem, what cases it occurs in, and what cases it doesn’t
occur in. In particular you will need to show the smallest test of your own that you were able to write that illustrates
the problem, so the cause can be narrowed down as much as possible before the TAs even start helping you.
You must also have used the gdb debugger, and be prepared to show the TAs how you attempted to debug your
program using it and what results you got.
• (If you don’t know anything about different UNIX shells or changing your shell you can ignore this item.) As the
UNIX tutorial says, the Grace systems and the account setup procedure assume you are using the tcsh shell. If you
change your shell on Grace to another one such as bash, or if you write any scripts using another shell, it’s up to you
to figure out that what you’re doing is correct. The TAs cannot help with this in office hours, and we can not make any
allowances in grading for errors caused by using a shell other than tcsh. We recommend using tcsh for everything
in this course. (There is nothing you have to do in this course where the shell version makes any difference.)
C Academic integrity
Please carefully read the academic honesty section of the syllabus. Any evidence of impermissible cooperation on
projects, use of disallowed materials or resources, publicly providing others access to your project code online, or unauthorized use of computer accounts, will be submitted to the Office of Student Conduct, which could result in an XF for the
course, or suspension or expulsion from the University. Be sure you understand what you are and what you are not permitted to do in regards to academic integrity when it comes to projects. These policies apply to all students, and the Student
Honor Council does not consider lack of knowledge of the policies to be a defense for violating them. More information is
in the course syllabus – please review it now.
The academic integrity requirements also apply to any test data for projects, which must be your own original work.
Exchanging test data or working together to write test cases is also prohibited.
© 2023 L. Herman; all rights reserved 10