AMD Typewriter x86 사용자 설명서

22007E/0—November 1999

;IN:EAX = dividend
;OUT:EAX = remainder
CDQ

   

;Sign extend into EDX

AND

EDX, (2^n–1)

;Mask correction (abs(divison)–1)

ADD

EAX, EDX

;Apply pre-correction

AND

EAX, (2^n–1)

;Mask out remainder (abs(divison)–1)

SUB

EAX, EDX

;Apply pre-correction, if necessary

MOV

[remainder], EAX

A 32-bit integer multiply by a constant has a latency of five
cycles. Therefore, use alternative code when multiplying by
certain constants. In addition, because there is just one
multiply unit, the replacement code may provide better
throughput.

The following code samples are designed such that the original
source also receives the final result. Other sequences are
possible if the result is in a different register. Adds have been
favored over shifts to keep code size small. Generally, there is a
fast replacement if the constant has very few 1 bits in binary.

More constants are found in the file multiply_by_constants.txt
located in the same directory where this document is located in
the SDK.

by 2:

ADD 

REG1, REG1 

;1 cycle

by 3:

LEA 

REG1, [REG1*2+REG1]

;2 cycles

by 4:

SHL 

REG1, 2

;1 cycle

by 5:

LEA 

REG1, [REG1*4+REG1]     ;2 cycles 

by 6:

LEA 

REG2, [REG1*4+REG1]     ;3 cycles

ADD 

REG1, REG2 

by 7: 

MOV

REG2, REG1

;2 cycles

SHL

REG1, 3

SUB

REG1, REG2

   
by 8: 

SHL 

REG1, 3              

;1 cycle

by 9: 

LEA 

REG1, [REG1*8+REG1]     ;2 cycles

by 10: 

LEA 

REG2, [REG1*8+REG1]     ;3 cycles

ADD 

REG1, REG2