AI 302

Embedded Systems
SPRING 2023

Lecture 06 – MIPS Floating Point Arithmetic

Instructor: Dr. Tarek Abdul Hamid
The World is Not Just Integers
 Programming languages support numbers with fraction
 Called floating-point numbers
 Examples:
3.14159265… (π)
2.71828… (e)
0.000000001 or 1.0 × 10–9 (seconds in a nanosecond)
86,400,000,000,000 or 8.64 × 1013 (nanoseconds in a day)
last number is a large integer that cannot fit in a 32-bit integer
 We use a scientific notation to represent
 Very small numbers (e.g. 1.0 × 10–9)
 Very large numbers (e.g. 8.64 × 1013)
 Scientific notation: ± d . f1f2f3f4 … × 10 ± e1e2e3

2 Embedded Systems Dr. Tarek Abdul Hamid

Floating-Point Numbers
 Examples of floating-point numbers in base 10 …
 5.341×103 , 0.05341×105 , –2.013×10–1 , –201.3×10–3
decimal point
 Examples of floating-point numbers in base 2 …
 1.00101×223 , 0.0100101×225 , –1.101101×2–3 , –1101.101×2–6
binary
 Exponents are kept in decimal for clarity point
 The binary number (1101.101)2 = 23+22+20+2–1+2–3 = 13.625
 Floating-point numbers should be normalized
 Exactly one non-zero digit should appear before the point
 In a decimal number, this digit can be from 1 to 9
 In a binary number, this digit should be 1
 Normalized FP Numbers: 5.341×103 and –1.101101×2–3
 NOT Normalized: 0.05341×105 and –1101.101×2–6

3 Embedded Systems Dr. Tarek Abdul Hamid

Floating-Point Representation
 A floating-point number is represented by the triple
 S is the Sign bit (0 is positive and 1 is negative)
 Representation is called sign and magnitude
 E is the Exponent field (signed)
 Very large numbers have large positive exponents
 Very small close-to-zero numbers have negative exponents
 More bits in exponent field increases range of values
 F is the Fraction field (fraction after binary point)
 More bits in fraction field improves the precision of FP numbers

S Exponent Fraction

Value of a floating-point number = (-1)S × val(F) × 2val(E)

4 Embedded Systems Dr. Tarek Abdul Hamid
IEEE 754 Floating-Point Standard
 Found in virtually every computer invented since 1980
 Simplified porting of floating-point numbers
 Unified the development of floating-point algorithms
 Increased the accuracy of floating-point numbers
 Single Precision Floating Point Numbers (32 bits)
 1-bit sign + 8-bit exponent + 23-bit fraction
S Exponent8 Fraction23

 Double Precision Floating Point Numbers (64 bits)

 1-bit sign + 11-bit exponent + 52-bit fraction
S Exponent11 Fraction52
(continued)
5 Embedded Systems Dr. Tarek Abdul Hamid
Biased Exponent Representation
 How to represent a signed exponent? Choices are …
 Sign + magnitude representation for the exponent
 Two’s complement representation
 Biased representation
 IEEE 754 uses biased representation for the exponent
 Value of exponent = val(E) = E – Bias (Bias is a constant)
 Recall that exponent field is 8 bits for single precision
 E can be in the range 0 to 255
 E = 0 and E = 255 are reserved for special use (discussed later)
 E = 1 to 254 are used for normalized floating point numbers
 Bias = 127 (half of 254), val(E) = E – 127
 val(E=1) = –126, val(E=127) = 0, val(E=254) = 127

6 Embedded Systems Dr. Tarek Abdul Hamid

Biased Exponent – Cont’d

 For double precision, exponent field is 11 bits

 E can be in the range 0 to 2047
 E = 0 and E = 2047 are reserved for special use
 E = 1 to 2046 are used for normalized floating point numbers
 Bias = 1023 (half of 2046), val(E) = E – 1023
 val(E=1) = –1022, val(E=1023) = 0, val(E=2046) = 1023
 Value of a Normalized Floating Point Number is

(–1)S × (1.F)2 × 2E – Bias

(–1)S × (1.f1f2f3f4 …)2 × 2E – Bias
(–1)S × (1 + f1×2-1 + f2×2-2 + f3×2-3 + f4×2-4 …)2 × 2E – Bias

7 Embedded Systems Dr. Tarek Abdul Hamid

Examples of Single Precision Float
 What is the decimal value of this Single Precision float?

10111110001000000000000000000000

 Solution:
 Sign = 1 is negative
 Exponent = (01111100)2 = 124, E – bias = 124 – 127 = –3
 Significand = (1.0100 … 0)2 = 1 + 2-2 = 1.25 (1. is implicit)
 Value in decimal = –1.25 × 2–3 = –0.15625
 What is the decimal value of?

01000001001001100000000000000000

 Solution: implicit
 Value in decimal = +(1.01001100 … 0)2 × 2130–127
= (1.01001100 … 0)2 × 23 = (1010.01100 … 0)2 = 10.375
8 Embedded Systems Dr. Tarek Abdul Hamid
Examples of Double Precision Float
 What is the decimal value of this Double Precision float ?

01000000010100101010000000000000
00000000000000000000000000000000
 Solution:
 Value of exponent = (10000000101)2 – Bias = 1029 – 1023 = 6
 Value of double float = (1.00101010 … 0)2 × 26 (1. is implicit)
= (1001010.10 … 0)2
= 74.5
 What is the decimal value of ?

10111111100010000000000000000000
00000000000000000000000000000000

 Do it yourself! (answer should be –1.5 × 2–7 = –0.01171875)

9 Embedded Systems Dr. Tarek Abdul Hamid

Converting FP Decimal to Binary
 Convert –0.8125 to binary in single and double precision
 Solution:
 Fraction bits can be obtained using multiplication by 2
 0.8125 × 2 = 1.625
 0.625 × 2 = 1.25
 0.25 × 2 = 0.5 0.8125 = (0.1101)2 = ½ + ¼ + 1/16 = 13/16
 0.5 × 2 = 1.0
 Stop when fractional part is 0
 Fraction = (0.1101)2 = (1.101)2 × 2 –1 (Normalized)
 Exponent = –1 + Bias = 126 (single precision) and 1022 (double)
Single
10111111010100000000000000000000
Precision
10111111111010100000000000000000 Double
Precision
00000000000000000000000000000000
10 Embedded Systems Dr. Tarek Abdul Hamid
MIPS Floating Point Coprocessor
 Called Coprocessor 1 or the Floating Point Unit (FPU)
 32 separate floating point registers: $f0, $f1, …, $f31
 FP registers are 32 bits for single precision numbers
 Even-odd register pair form a double precision register
 Use the even number for double precision registers
 $f0, $f2, $f4, …, $f30 are used for double precision
 Separate FP instructions for single/double precision
 Single precision: add.s, sub.s, mul.s, div.s (.s extension)
 Double precision: add.d, sub.d, mul.d, div.d (.d extension)
 FP instructions are more complex than the integer ones
 Take more cycles to execute

11 Embedded Systems Dr. Tarek Abdul Hamid

The MIPS Processor
...

4 bytes per word Memory

Up to 232 bytes = 230 words
...

EIU $0 Execution & FPU F0 Floating

32 General $1 Integer Unit F1 Point Unit
Purpose $2 (Main proc) F2 (Coproc 1) 32 Floating-Point
Registers Registers
$31 F31
Arithmetic & Integer FP
ALU
Logic Unit mul/div Arith
Floating-Point
Arithmetic Unit
Hi Lo
TMU BadVaddr Trap &
Status Memory Unit
Cause (Coproc 0)
Integer EPC
Multiplier/Divider

12 Embedded Systems Dr. Tarek Abdul Hamid

FP Arithmetic Instructions
Instruction Meaning Format
add.s fd, fs, ft (fd) = (fs) + (ft) 0x11 0 ft5 fs5 fd5 0
add.d fd, fs, ft (fd) = (fs) + (ft) 0x11 1 ft5 fs5 fd5 0
sub.s fd, fs, ft (fd) = (fs) – (ft) 0x11 0 ft5 fs5 fd5 1
sub.d fd, fs, ft (fd) = (fs) – (ft) 0x11 1 ft5 fs5 fd5 1
mul.s fd, fs, ft (fd) = (fs) × (ft) 0x11 0 ft5 fs5 fd5 2
mul.d fd, fs, ft (fd) = (fs) × (ft) 0x11 1 ft5 fs5 fd5 2
div.s fd, fs, ft (fd) = (fs) / (ft) 0x11 0 ft5 fs5 fd5 3
div.d fd, fs, ft (fd) = (fs) / (ft) 0x11 1 ft5 fs5 fd5 3
sqrt.s fd, fs (fd) = sqrt (fs) 0x11 0 0 fs5 fd5 4
sqrt.d fd, fs (fd) = sqrt (fs) 0x11 1 0 fs5 fd5 4
abs.s fd, fs (fd) = abs (fs) 0x11 0 0 fs5 fd5 5
abs.d fd, fs (fd) = abs (fs) 0x11 1 0 fs5 fd5 5
neg.s fd, fs (fd) = – (fs) 0x11 0 0 fs5 fd5 7
neg.d fd, fs (fd) = – (fs) 0x11 1 0 fs5 fd5 7
FP Load/Store Instructions
 Separate floating point load/store instructions
 lwc1: load word coprocessor 1
General purpose
 ldc1: load double coprocessor 1
register is used as
 swc1: store word coprocessor 1 the base register
 sdc1: store double coprocessor 1
Instruction Meaning Format
lwc1 $f2, 40($t0) ($f2) = Mem[($t0)+40] 0x31 $t0 $f2 im16 = 40
ldc1 $f2, 40($t0) ($f2) = Mem[($t0)+40] 0x35 $t0 $f2 im16 = 40
swc1 $f2, 40($t0) Mem[($t0)+40] = ($f2) 0x39 $t0 $f2 im16 = 40
sdc1 $f2, 40($t0) Mem[($t0)+40] = ($f2) 0x3d $t0 $f2 im16 = 40

 Better names can be used for the above instructions

 l.s = lwc1 (load FP single), l.d = ldc1 (load FP double)
 s.s = swc1 (store FP single), s.d = sdc1 (store FP double)
FP Data Movement Instructions
 Moving data between general purpose and FP registers
 mfc1: move from coprocessor 1 (to general purpose register)
 mtc1: move to coprocessor 1 (from general purpose register)

 Moving data between FP registers

 mov.s: move single precision float
 mov.d: move double precision float = even/odd pair of registers

Instruction Meaning Format

mfc1 $t0, $f2 ($t0) = ($f2) 0x11 0 $t0 $f2 0 0
mtc1 $t0, $f2 ($f2) = ($t0) 0x11 4 $t0 $f2 0 0
mov.s $f4, $f2 ($f4) = ($f2) 0x11 0 0 $f2 $f4 6
mov.d $f4, $f2 ($f4) = ($f2) 0x11 1 0 $f2 $f4 6
FP Convert Instructions
 Convert instruction: cvt.x.y
 Convert to destination format x from source format y
 Supported formats
 Single precision float = .s (single precision float in FP register)
 Double precision float = .d (double float in even-odd FP register)
 Signed integer word = .w (signed integer in FP register)

Instruction Meaning Format

cvt.s.w fd, fs to single from integer 0x11 0 0 fs5 fd5 0x20
cvt.s.d fd, fs to single from double 0x11 1 0 fs5 fd5 0x20
cvt.d.w fd, fs to double from integer 0x11 0 0 fs5 fd5 0x21
cvt.d.s fd, fs to double from single 0x11 1 0 fs5 fd5 0x21
cvt.w.s fd, fs to integer from single 0x11 0 0 fs5 fd5 0x24
cvt.w.d fd, fs to integer from double 0x11 1 0 fs5 fd5 0x24
FP Compare and Branch Instructions
 FP unit (co-processor 1) has a condition flag
 Set to 0 (false) or 1 (true) by any comparison instruction
 Three comparisons: equal, less than, less than or equal
 Two branch instructions based on the condition flag

Instruction Meaning Format

c.eq.s fs, ft cflag = ((fs) == (ft)) 0x11 0 ft5 fs5 0 0x32
c.eq.d fs, ft cflag = ((fs) == (ft)) 0x11 1 ft5 fs5 0 0x32
c.lt.s fs, ft cflag = ((fs) <= (ft)) 0x11 0 ft5 fs5 0 0x3c
c.lt.d fs, ft cflag = ((fs) <= (ft)) 0x11 1 ft5 fs5 0 0x3c
c.le.s fs, ft cflag = ((fs) <= (ft)) 0x11 0 ft5 fs5 0 0x3e
c.le.d fs, ft cflag = ((fs) <= (ft)) 0x11 1 ft5 fs5 0 0x3e
bc1f Label branch if (cflag == 0) 0x11 8 0 im16
bc1t Label branch if (cflag == 1) 0x11 8 1 im16
FP Data Directives
 .FLOAT Directive
 Stores the listed values as single-precision floating point

 .DOUBLE Directive
 Stores the listed values as double-precision floating point

 Examples
 var1: .FLOAT 12.3, -0.1
 var2: .DOUBLE 1.5e-10
 pi: .DOUBLE 3.1415926535897924

18 Embedded Systems Dr. Tarek Abdul Hamid

FP Instructions in MIPS
 FP hardware is coprocessor 1
 Adjunct processor that extends the ISA
 Separate FP registers
 32 single-precision: $f0, $f1, … $f31
 Paired for double-precision: $f0/$f1, $f2/$f3, …
 Release 2 of MIPs ISA supports 32 × 64-bit FP reg’s

 FP instructions operate only on FP registers

 Programs generally don’t do integer ops on FP data, or vice
versa
 More registers with minimal code-size impact
 FP load and store instructions
 lwc1, ldc1, swc1, sdc1
19
e.g., ldc1 Systems
 Embedded $f8, 32($sp) Dr. Tarek Abdul Hamid
FP Instructions in MIPS
 Single-precision arithmetic
 add.s, sub.s, mul.s, div.s
 e.g., add.s $f0, $f1, $f6
 Double-precision arithmetic
 add.d, sub.d, mul.d, div.d
 e.g., mul.d $f4, $f4, $f6

 Single- and double-precision comparison

 c.xx.s, c.xx.d (xx is eq, lt, le, …)
 Sets or clears FP condition-code bit
 e.g. c.lt.s $f3, $f4

 Branch on FP condition code true or false

 bc1t, bc1f
 e.g., bc1t TargetLabel
20 Embedded Systems Dr. Tarek Abdul Hamid
Exercise 6 :
Convert the following C fragment to equivalent MIPS assembly language

F = 8x2 + 9x + 15

“Assume that the variables x and F are assigned to registers $f0 and $f2 respectively”

21 Embedded Systems Dr. Tarek Abdul Hamid

Exercise 6 :
addi $to, $zero, 8 # to = 8
mtc1 $t0, $f10 # f10 = 8
cvt.s.w $f11, $f10 # f11 = (8)sp

mul.s $f11, $f11, $f0 # f11 = 8x

addi $t1, $zero, 9 # t1 = 9

mtc1 $t1, $f12 # f12 = 9
cvt.s.w $f13, $f12 # f13 = (9)sp

add.s $f2,$f11,$f13 # F = 8x + 9
mul.s $f2,$f2,$f0 # F = (8x+9) x = 8x2 + 9x

addi $t2, $zero, 15 # t2 = 15

mtc1 $t2, $f14 # f14 = 15
cvt.s.w $f15, $f14 # f15 = (15)sp

add.s $f2,$f2,$f15 # F = 8ax^2 + 9x + 15

22 Embedded Systems Dr. Tarek Abdul Hamid