Real Arithmetic
Computer Organization and Assembly Languages p g z y g g
Yung-Yu Chuang
Fractional binary numbers
2i–1 2i
2 4
• • •
bi bi–1 • • • b2 b1 b0 . b–1 b–2 b–3 • • • b–j 1
• • • 1/2
1/4 1/8
• Representation p
2–j– Bits to right of “binary point” represent fractional powers of 2 i bb 22k
Binary real numbers
• Binary real to decimal real
• Decimal real to binary real
4.5625 = 100.1001
2Fractional binary numbers examples
•Value Representation
5 3/4 101 11
5-3/4 101.112
2-7/8 10.1112
63/64 0 111111
63/64 0.1111112
•Value Representation
1/3 0 0101010101[01]
1/3 0.0101010101[01]…2
1/5 0.001100110011[0011]…2 1/10 0 0001100110011[0011]…2 1/10 0.0001100110011[0011]…2
Fixed-point numbers
sign integer part fractional part
radix point
0 000 0000 0000 0110 0110 0000 0000 0000 = 110.011
• only 2 only 2 to 2
16to 2
-16Not flexible, not adaptive to applications
• Fast computation, just integer operations. Fast computation, just integer operations.
It is often a good way to speed up in this way If you know the working range beforehand.
If you know the working range beforehand.
IEEE floating point
• IEEE Standard 754
E t bli h d i 1985 if t d d f fl ti – Established in 1985 as uniform standard for floating
point arithmetic
• Before that many idiosyncratic formats
• Before that, many idiosyncratic formats – Supported by all major CPUs
Driven by Numerical Concerns
• Driven by Numerical Concerns
– Nice standards for rounding, overflow, underflow H d t k g f t
– Hard to make go fast
• Numerical analysts predominated over hardware types in defining standard
types in defining standard
IEEE floating point format
• IEEE defines two formats with different precisions: single and double
precisions: single and double
23.85 = 10111.110110
2=1.0111110110x2
4h
0 100 0001 1 011 1110 1100 1100 1100 1100 e = 127+4=83h
0 100 0001 1 011 1110 1100 1100 1100 1100
IEEE floating point format
special values special values
IEEE double precision
IEEE double precision
Denormalized numbers
• Number smaller than 1.0x2
-126can’t be
presented by a single with normalized form presented by a single with normalized form.
However, we can represent it with denormalized format
denormalized format.
• 1.0000..00x2
-126the least “normalized” number
• 0.1111..11x2
-126the largest “denormalized”
number
• 1.001x2
-129=0.001001x2
-126Summary of Real Number Encodings
+
NaN NaN
+
0
+Denorm +Normalized -Denorm
-Normalized
NaN 0 +00
(3.14+1e20)-1e20=0
3.14+(1e20-1e20)=3.14
IA-32 floating point architecture
• Original 8086 only has integers. It is possible to simulate real arithmetic using software but it simulate real arithmetic using software, but it is slow.
8087 fl ti i t ( d 80287 80387)
• 8087 floating-point processor (and 80287, 80387) was sold separately at early time.
• Since 80486, FPU (floating-point unit) was
integrated into CPU.
FPU data types
• Three floating-point types
FPU data types
• Four integer types
FPU registers
• Data register
C l i
• Control register
• Status register
• Tag register
Data registers
• Load: push, TOP--
• Store: pop TOP++
79 0
• Store: pop, TOP++ R0
• Instructions access the stack using ST(i)
R1
R2 ST(0) 010
TOP
g ( ) relative to TOP
• If TOP=0 and push, TOP
R3 R4
ST(1) ST(2)
p wraps to R7
• If TOP=7 and pop, TOP
R5 R6
( )
wraps to R0
• When overwriting occurs, t ti
R7
generate an exception
• Real values are transferred to and from memory and stored in 10-byte temporary format When storing stored in 10-byte temporary format. When storing, convert back to integer, long, real, long real.
Postfix expression
• (5*6)-4 → 5 6 * 4 -
6 4
5 5
5 6
30
*
30 4
26 -
Special-purpose registers
Special-purpose registers
• Last data pointer stores the memory address of the operand for the last non control instruction the operand for the last non-control instruction.
Last instruction pointer stored the address of the last non control instruction Both are 48 the last non-control instruction. Both are 48 bits, 32 for offset, 16 for segment selector.
1 1 0 1 1 1 1 0 1 1
Control register
Initial 037Fh
for compatibility only for compatibility only
The instruction FINIT will initialize it to 037Fh.
Rounding
• FPU attempts to round an infinitely accurate result from a floating point calculation
result from a floating-point calculation
– Round to nearest even: round toward to the closest one; if both are equally close round to the even one one; if both are equally close, round to the even one – Round down: round toward to -∞
Round up: round toward to + – Round up: round toward to +∞
– Truncate: round toward to zero
E l
• Example
– suppose 3 fractional bits can be stored, and a l l t d l l 1 0111
calculated value equals +1.0111.
– rounding up by adding .0001 produces 1.100
di d b bt ti 0001 d 1 011 – rounding down by subtracting .0001 produces 1.011
Rounding
method original value rounded value Round to nearest even 1.0111 1.100
Round down 1.0111 1.011
Round up 1.0111 1.100
Truncate 1 0111 1 011
Truncate 1.0111 1.011
methodet od original valueo g al value rounded valueou ded value Round to nearest even -1.0111 -1.100
Round down -1.0111 -1.100
Round down 1.0111 1.100
Round up -1.0111 -1.011
T t 1 0111 1 011
Truncate -1.0111 -1.011
Floating-Point Exceptions
• Six types of exception conditions
#I I lid ti – #I: Invalid operation – #Z: Divide by zero
#D D li d d
detect before execution – #D: Denormalized operand
– #O: Numeric overflow
# d fl d t t ft ti
– #U: Numeric underflow – #P: Inexact precision
detect after execution
• Each has a corresponding mask bit
– if set when an exception occurs, the exception is handled automatically by FPU
– if clear when an exception occurs, a software i h dl i i k d
exception handler is invoked
Status register
C3-C0: condition bits after comparisons
FPU data types
.data
bigVal REAL10 1 212342342234234243E+864 bigVal REAL10 1.212342342234234243E+864 .code
fld bigVal fld bigVal
FPU instruction set
• Instruction mnemonics begin with letter F
S d l id ifi d f
• Second letter identifies data type of memory operand
– B = bcd – I = integer
– no letter: floating point
• Examples
– FBLD load binary coded decimal – FISTP store integer and pop stackg p p
– FMUL multiply floating-point operands
FPU instruction set
• Fop {destination}, {source}
Operands
• Operands
– zero, one, or two
• faddfadd
• fadd [a]
• fadd st, st(1)
– no immediate operands
– no general-purpose registers (EAX, EBX, ...) (FSTSW is the only exception which stores FPU status word is the only exception which stores FPU status word to AX)
– destination must be a stack registerdestination must be a stack register
– integers must be loaded from memory onto the stack and converted to floating-point before being used in
l l i
calculations
Classic stack (0-operand)
• ST(0) as source, ST(1) as destination. Result is
stored at ST(1) and ST(0) is popped leaving the
stored at ST(1) and ST(0) is popped, leaving the
result on the top. (with 0 operand,
fadd=faddp)
Memory operand (1-operand)
• ST(0) as the implied destination. The second operand is from memory
operand is from memory.
Register operands (2-operand)
• Register: operands are FP data registers, one must be ST
must be ST.
• Register pop: the same as register with a ST g p p g
pop afterwards.
Example: evaluating an expression
Load
FLDPI stores π
FLDL2T stores log2(10) FLDL2E stores log2(e) FLDLG2 stores log10(2) FLDLN2 stores ln(2)( )
load
.data
array REAL8 10 DUP(?)y ( ) .code
fld array ; direct
fld [array+16] ; direct-offset
fld REAL8 PTR[esi] ; indirect
fld array[esi] ; indexed
fld array[esi] ; indexed
fld array[esi*8] ; indexed, scaled
fld REAL8 PTR[ebx+esi]; base-index
fld array[ebx+esi] ; base-index-displacement
Store
Store
fst dblOne ; 200.0
f t dblT 200 0
fst dblTwo ; 200.0
fstp dblThree ; 200.0
fstp dblFour ; 32.0
Arithmetic instructions
FCHS ; change sign of ST FABS ; ST=|ST|
Floating-Point add
• FADD
dd t d ti ti – adds source to destination
– No-operand version pops the FPU stack after addition
stack after addition
• Examples:
Floating-Point subtract
• FSUB
bt t f d ti ti
– subtracts source from destination.
– No-operand version pops the FPU stack after subtracting
stack after subtracting
• Example:
fsub mySingley g ; ST -= mySingley g
fsub array[edi*8] ; ST -= array[edi*8]
Floating-point multiply/divide
• FMUL
M lti li b d ti ti
– Multiplies source by destination, stores product in destination
• FDIV
– Divides destination by source, then pops the stack
Miscellaneous instructions
.data
x REAL4 2.75. 5 five REAL4 5.2 .code
.code
fld five ; ST0=5.2
fld x ; ST0=2 75, ST1=5 2
fld x ; ST0 2.75, ST1 5.2
fscale ; ST0=2.75*32=88
; ST1=5 2
; ST1=5.2
Example: compute distance
; compute D=sqrt(x^2+y^2)
fld x ; load x
fld x ; load x
fld st(0) ; duplicate x
fmul ; x*x
fmul ; x*x
fld y ; load y
fld y ; load y
fld st(0) ; duplicate y
f l *
fmul ; y*y
f dd * *
fadd ; x*x+y*y
fsqrt fst D
Example: expression
; expression:valD = –valA + (valB * valC).
data .data
valA REAL8 1.5 valB REAL8 2 5 valB REAL8 2.5 valC REAL8 3.0
valD REAL8 ? ; will be +6.0 valD REAL8 ? ; will be +6.0 .code
fld valA ; ST(0) = valA fld valA ; ST(0) valA
fchs ; change sign of ST(0) fld valB d a ; load valB into ST(0); oad a to ( ) fmul valC ; ST(0) *= valC
fadd ; ST(0) += ST(1); ( ) ( )
Example: array sum
.data N = 20 N 20
array REAL8 N DUP(1.0) sum REAL8 0.0
.code
mov ecx, N
mov esi, OFFSET array
fldz ; ST0 = 0
lp: fadd REAL8 PTR [esi]; ST0 += *(esi)
add esi, 8 ; move to next double loop lp
fstp sum ; store result
Comparisons
Comparisons
• The above instructions change FPU’s status register of FPU and the following instructions register of FPU and the following instructions are used to transfer them to CPU.
• SAHF copies C into carry C into parity and C
• SAHF copies C
0into carry, C
2into parity and C
3to zero. Since the sign and overflow flags are
not set use conditional jumps for unsigned
not set, use conditional jumps for unsigned
integers (ja, jae, jb, jbe, je, jz).
Comparisons
Branching after FCOM
• Required steps:
1 Use the FSTSW instruction to move the FPU status 1. Use the FSTSW instruction to move the FPU status
word into AX.
2 Use the SAHF instruction to copy AH into the 2. Use the SAHF instruction to copy AH into the
EFLAGS register.
3 Use JA JB etc to do the branching 3. Use JA, JB, etc to do the branching.
• Pentium Pro supports two new comparison instructions that directly modify CPU’s FLAGS instructions that directly modify CPU s FLAGS.
FCOMI ST(0), src ; src=STn FCOMIP ST(0), src( ),
Example
fcomi ST(0), ST(1) jnb Label1
Example: comparison
.data
x REAL8 1.0 y REAL8 2.0 .code
; if (x>y) return 1 else return 0
; if (x>y) return 1 else return 0
fld x ; ST0 = x
fcomp y ; compare ST0 and y
fstsw ax ; move C bits into FLAGS sahf
jna else part ; if x not above y,
jna else_part ; if x not above y, ...
then_part:
mov eax, 1 jmp end_if else_part:
mov eax, 0,
Example: comparison
.data
x REAL8 1.0 y REAL8 2.0 .code
; if (x>y) return 1 else return 0
; if (x>y) return 1 else return 0
fld y ; ST0 = y
fld x ; ST0 = x ST1 = y
fcomi ST(0), ST(1)
jna else part ; if x not above y,
jna else_part ; if x not above y, ...
then_part:
mov eax, 1 jmp end_if else_part:
mov eax, 0, end_if:
Comparing for equality
• Not to compare floating-point values directly because of precision limit For example
because of precision limit. For example, sqrt(2.0)*sqrt(2.0) != 2.0
instruction FPU stack
fld two ST(0): +2.0000000E+000 fsqrt ST(0): +1.4142135+000 fm l ST(0) ST(0) ST(0) +2 0000000E+000 fmul ST(0), ST(0) ST(0): +2.0000000E+000 fsub two ST(0): +4.4408921E-016
Comparing for equality
• Calculate the absolute value of the difference between two floating point values
between two floating-point values
.data
epsilon REAL8 1.0E-12 ; difference value val2 REAL8 0.0 ; value to compare
val3 REAL8 1.001E-13 ; considered equal to val2 .code
; if( val2 == val3 ), display "Values are equal".
fld epsilon fld epsilon fld val2
fsub val3 fabs
fabs
fcomi ST(0),ST(1) ja skip
mWrite <"Values are equal",0dh,0ah>