Project 1 Floating-Point Operations without Floating-Point Solution

Computer Organization & Assembly Language




Check the Due Date on the CourseWeb







The purpose of this project is for you to practice writing assembly language to interact with input/output hardware. The hardware for this project is a very basic oating-point calculator as shown below:

















































This tool can be found in q88 calculator.zip located in the CourseWeb under this project. Extract all les to your [..]/mars4 5/mars/tools directory. If you extract all les to the right directory, when you run the MARS program, you should see "Q8.8 Calculator (Register) V0.1" under the "Tools" menu.







Introduction to the Q (Number Format)




In some embedded systems, oating-point operations are not included in their implementation to keep the cost down. In other words, those embedded systems only have integer operations. To perform oating-point operations in such embedded systems, programmer must implement his/her own representation of oating-point numbers and perform calculation based on implementations.




One simple oating-point representation is called the Q number format. Q is a xed point number format where the number of bits of integer and fractions are speci ed. For example, a Q8.8 number has 8 integer bit and 8 fractional bit. In this project, we will only focus on Q8.8 format.




A Q8.8 format is a 16-bit number. As mentioned earlier, it consists of 8 integer bit and 8 fractional bit. The 8 integer bit is at the top 8 bits (bit 8 to bit 15) and the fractional bit is at the bottom 8 bit (bit 0 to bit 7) as shown below:




15 8 7 0







integer fractional













1



A Q8.8 format uses Two’s Complement format to represent negative value. To get a Q8.8 represen-tation of a oating-point number, simply multiply the oating-point number by 256 and rounded to the nearest integer. For example, the Q8.8 representation of the number 3.25 is 3:25 256 = 832. Recall that 832 in 16-bit two’s complement format is 0000001101000000 which is Q8.8 represen-tation of 3.25. The Q8.8 format of -3.25 can be easily obtained as discussed in class by simply ip 3.25 bit-by-bit and add 1. Thus, the Q8.8 of -3.25 is 1111110011000000. To convert a Q8.8 format back to a oating-point number, simply divide the integer value of Q8.8 by 256. For ex-ample, suppose a Q8.8 number is 0011001100110000 which is 13104, the value is oating-point is 13104=256 = 51:1875. Simply put, if x is a oating-point number, x in Q8.8 format is x 256:0 rounded to the closest integer.




Q8.8 Addition




Note that (x 256:0) + (y 256:0) = (x + y) 256:0. In other words, to perform A + B where A and B are oating-point numbers in Q8.8 format, simply perform integer addition A + B. The result is a oating-point number in Q8.8 format. For example, consider two oating-point numbers is Q8.8 format, 1111110011000000 (-3.25) and 0011001100110000 (51.1857). Addition in binary is as follows:




1111110011000000 + 0011001100110000 = 0010111111110000




Note that 0010111111110000 is 12272 in decimal and 12272=256 = 47:9375 which is the result of (-3.25) + 51.1875.




Q8.8 Subtraction




Similar to Q8.8 addition, (x 256:0) (y 256:0) = (x y) 256:0. Thus, to perform A B where




A and B are oating-point number in Q8.8 format, simply perform integer subtraction A B. The




result is a oating-point number in Q8.8 format.




Q8.8 Multiplication




For multiplication, since (x 256:0) (y 256:0) = (x y) (256:0 256:0), a result of a multiplication is in Q16.16 format. Note that an 8-bit number multiplied by an 8-bit number results in a 16-bit number, and (256:0 256:0 = 216. Thus, to perform A B where A and B are oating-point numbers in Q8.8 format, simply perform integer multiplication A B. However, the result is in Q16.16 format as discussed earlier. So, we need to convert a Q16.16 back to Q8.8. For simplicity, simply shift the result to the right by 8 (same as divided by 256) to get Q16.8 format. For this project, we will only test with small numbers to ensure that the result of a multiplication is still in the range of Q8.8 format representation.




For this project, you are NOT ALLOWED to use any multiplication instructions. You must implement your multiplication operation using basic instructions such as addition, shifting, and bit-wise operations. Note that you only need to implement unsigned multiplication. Signed multiplication can be easily implemented by; (1) converting any negative operands to positive, (2) performing unsigned multiplication, and (3) converting the result based on signs of operands. An unsigned multiplication method in binary will be discussed in a lecture.






















2



Q8.8 Division




For division, since (x 256:0)=(y 256:0) = x=y, a result of a division (quotient) is in Q8.0 format and the remainder (x mod y) is also in Q8.0 format. Thus, to perform A=B where A and B are oating-point numbers in Q8.8 format, simply perform integer division A=B. Note that an integer division generally gives a quotient and a remainder. As discussed earlier, the quotient is in Q8.0 format (no fraction) and the remainder is in Q8.0 format. There are multiple ways to get the quotient in Q8.8 format. One simple way is to convert the dividend to Q8.16 by shifting left 8 times (multiply by 256). Then, perform integer division as usual to get a quotient in Q8.8 format. However, according to the division algorithm discussed in class, this work around requires us to shift the divisor left 24 times which results in a 40-bit number. Unfortunately, a register in MIPS is only 32-bit wide. One solution is to perform A=B to get a quotient1 and a remainder. Then perform (remainder << 8)=B to get quotient2. The quotient in Q8.8 format is simply (quotient1 << 8)jquotient2.




For this project, you are NOT ALLOWED to use any division instructions. You must implement your multiplication operation using basic instruction such as addition, subtraction, comparisons, shifting, and bit-wise operation. Similar to multiplication, you only need to implement unsigned division and extend it to support signed division. The unsigned division method in binary will be discussed in a lecture.




Q8.8 Square Root




A square root of a Q8.8 number results in a Q4.4 number. We can use a simple trick by converting a Q8.8 number into Q8.16 to get a Q4.8 result. For this project, you only need to calculate the square root of the absolute value of the operand A only. The absolute value is required since a square root of a negative number is an imaginary number. A square root method in binary will be discussed in a lecture.




Q8.8 Calculator Hardware




The Q8.8 calculator hardware allows user to enter two operands A and B in oating-point format.




It will convert them into Q8.8 format and display some information as follows:




A 256 (rounded): This is the value of the operand A multiplied by 256 and rounded to the closest integer




Approximate: This is the approximate value of the oating-point number in Q8.8 format. Note that not all oating-point value can be represented exactly by Q8.8 format. For example, 2:3 256 = 588:8 which must be rounded to 589 (00000010010011012). If we convert 589 back, we get 589=256:0 = 2:30078125.




Q8.8: This shows the given number in Q8.8 format (16-bit Two’s complement)




When a user click the \Calculate" button, the Q8.8 calculator will put the operand A in the register $a0 and put the operand B in register $a1 (both in Q8.8 16-bit Two’s complement format). Then the program will set the register $t9 to 1 and wait until the value of $t9 is back to 0 (by your program). Once it detects that $t9 is back to 0, it will display and check the results (given by your program). If a result is in an acceptable range, it will display in blue. Otherwise, your result will be displayed in red.







3



What you need to do is to write a program that calculate A + B, A B, A B, A=B, and

p




jAj. These results must by put in the following registers:




$v0 should contain the result of A + B and A B. Put the result of A + B into the lower 16-bit and A B into the upper 16-bit.




$v1 should contain the result of A B and A=B. Put the result of A B into the lower




16-bit and A=B into the upper 16-bit.

p




$a2 should contain the result of jAj.




Results must be put into three registers as shown below:




$v0
A − B
A + B
$v1








A / B
A x B












$a2


sqrt(|A|)









The Q8.8 calculator hardware also has a test feature. If a user press the \Run Test" button, the Q8.8 calculator will perform the following steps:




randomly generate numbers (in integer as well as oating-point) and put them as operand A and operand B,



set $t9 to 1,



wait until $t9 is changed to 0, check results,



Go back to step 1



It will keep running for roughly 6000 iterations (1000 iterations for integer operands and 5000 iterations for oating-point operands) and stop or until your program gives an incorrect result. You can tell the Q8.8 calculator to ignore a speci c result of calculation by checking its associated check box. In doing so, if a result of that speci c calculation is incorrect, it will keep running test. However, the incorrect result will be shown in red as usual.




To make your program compatible with this testing feature, the structure of your program should look like the following:







.text




wait: beq $t9, $zero, wait # Wait until $t9 is not zero (1)




Get operands A and B from $a0 and $a1



Calculate A + B and put the result in the lower 16-bit of $v0



Calculate A - B and put the result in the higher 16-bit of $v0



Calculate A * B and put the result in the lower 16-bit of $v1



Calculate A / B and put the result in the higher 16-bit of $v1



Calculate sqrt(|A|) and put the result in the lower 16-bit of $a2



add
$t9, $zero, $zero
#
Set $t9
back to
0
j
wait
#
Go back
to wait








During your calculation, you MUST NOT use the register $t9.




4
Requirements




The following are requirements that you must follow:




Your program should be named q88.asm




Since the Q8.8 Calculator uses registers $t9 to communicate between the program and the hardware, do not use $t9 for any other purposes.




No need to have functions in your program. In other words, do not use instructions jal and jr or stack pointer ($sp) unless you are really familiar with functions in assembly.




No memory reference instructions are allowed (no load and store instructions).




Your program should have only the text segment (.text) and no data segment (.data).




You are not allowed to use any multiplication, division, and square root instructions. You must implement your own signed multiplication, signed division, and square root algorithms by yourself.




You are NOT ALLOWED to use any oating-point instructions.




Grading Criteria




This project will be graded based on the functionality of your program. We will mainly test your program by simply click the \Run Test" button multiple times. If there is no error, you will get 100 points. If there are errors, points will be given according to the following:




Functionality
Points




Able to assemble
5
A + B
10
A B
10
positive positive
10
positive negative
5
negative positive
5
negative negative
5
positive / positive
10
positive / negative
5
negative / positive
5
negative / negative
5
square root
25







Submission




The due date of this project is stated on the CourseWeb. Late submissions will not be accepted.




You should submit the le q88.asm via CourseWeb.






















5