Message-Id: <m0zC8by-000S4wC@inti.gov.ar>
Comments: Authenticated sender is <salvador AT natacha DOT inti DOT gov DOT ar>
From: "Salvador Eduardo Tropea (SET)" <salvador AT inti DOT gov DOT ar>
Organization: INTI
To: djgpp AT delorie DOT com
Date: Thu, 27 Aug 1998 17:27:01 +0000
MIME-Version: 1.0
Content-type: text/plain; charset=US-ASCII
Content-transfer-encoding: 7BIT
Subject: Constraints
Precedence: bulk

Here are some notes about gcc constraints in x86 architecture. I want 
comments, corrections and additions.

1) x86 constraints:

1.1) Classes of registers:

NO_REGS
AREG             /* %eax */
DREG             /* %edx */
CREG             /* %ecx */
BREG             /* %edx */
AD_REGS          /* %eax/%edx for DImode */
Q_REGS           /* %eax %ebx %ecx %edx */
SIREG            /* %esi */
DIREG            /* %edi */
INDEX_REGS       /* %eax %ebx %ecx %edx %esi %edi %ebp */
GENERAL_REGS     /* %eax %ebx %ecx %edx %esi %edi %ebp %esp */
FP_TOP_REG       /* %st(0) */
FP_SECOND_REG	  /* %st(1) */
FLOAT_REGS
ALL_REGS
LIM_REG_CLASSES

1.2) Then the architecture constraints are:

'r' GENERAL_REGS
'q' Q_REGS
'f' FLOAT_REGS
't' FP_TOP_REG
'u' FP_SECOND_REG
'a' AREG
'b' BREG
'c' CREG
'd' DREG
'A' AD_REGS
'D' DIREG
'S' SIREG

2) The inmediate constraint:

This constraint is named 'i' but some times you can't use any inmediate
value and you are restricted to a range. If the number is outside the
range you must use a register. For this reason the following are defined
for x86:

'I' VALUE >= 0 && (VALUE) <= 31
'J' VALUE >= 0 && (VALUE) <= 63
'K' VALUE == 0xff
'L' VALUE == 0xffff
'M' VALUE >= 0 && (VALUE) <= 3
'N' VALUE >= 0 && (VALUE) <= 255
'O' VALUE >= 0 && (VALUE) <= 32

As it could look unclear here is an example:

asm("movl %0,%%eax" : : "Ib" (value));

Then if value is in [0..31] we will get:

movl $inmediate value,%eax

If value is outside:

movl $inmediate value,%ebx
movl %ebx,%eax

The comments say that the use is:

I is for non-DImode shifts.
J is for DImode shifts.
K and L are for an `andsi' optimization.
M is for shifts that can be executed by the "lea" opcode.

3) General notes:

* You can use a list of constraints, the first match will be used.

* The 'm' constraint means memory if you don't use it gcc will force the
use of a register, so if the value comes from a variable and goes to a
register you'll want to use 'm' to avoid loading a register and then
moving it to other. Example:

int pp=300;
...
asm("movl %0,%%eax" : : "q" (pp));

Could generate:

movl _pp,%ebx
movl %ebx,%eax

but asm("movl %0,%%eax" : : "qm" (pp));

Will generate:

movl _pp,%eax

* The size of the variables (0,1 ...) can be specified using 'k','w' and
'b'. The default is 'k'.

Example:
asm("movb %0,(%%esi)" : : "a" (10));

Will generate

movb $10,%al
movb %eax,(%esi)

Wrong of course, but:
asm("movb %b0,(%%esi)" : : "a" (10));

Generates

movb $10,%al
movb %al,(%esi)

SET
------------------------------------ 0 --------------------------------
Visit my home page: http://set-soft.home.ml.org/
or
http://www.geocities.com/SiliconValley/Vista/6552/
Salvador Eduardo Tropea (SET). (Electronics Engineer)
Alternative e-mail: set-soft AT usa DOT net set AT computer DOT org
ICQ: 2951574
Address: Curapaligue 2124, Caseros, 3 de Febrero
Buenos Aires, (1678), ARGENTINA
TE: +(541) 759 0013