Subversion Repositories openrisc

@c Copyright 1991, 1992, 1993, 1994, 1995, 1997, 1999, 2002

@c Free Software Foundation, Inc.

@c This is part of the GAS manual.

@c For copying conditions, see the file as.texinfo.

@ifset GENERIC

@page

@node Sparc-Dependent

@chapter SPARC Dependent Features

@end ifset

@ifclear GENERIC

@node Machine Dependencies

@chapter SPARC Dependent Features

@end ifclear

@cindex SPARC support

@menu

* Sparc-Opts::                  Options

* Sparc-Aligned-Data::          Option to enforce aligned data

* Sparc-Syntax::                Syntax

* Sparc-Float::                 Floating Point

* Sparc-Directives::            Sparc Machine Directives

@end menu

@node Sparc-Opts

@section Options

@cindex options for SPARC

@cindex SPARC options

@cindex architectures, SPARC

@cindex SPARC architectures

The SPARC chip family includes several successive versions, using the same

core instruction set, but including a few additional instructions at

each version.  There are exceptions to this however.  For details on what

instructions each variant supports, please see the chip's architecture

reference manual.

By default, @code{@value{AS}} assumes the core instruction set (SPARC

v6), but ``bumps'' the architecture level as needed: it switches to

successively higher architectures as it encounters instructions that

only exist in the higher levels.

If not configured for SPARC v9 (@code{sparc64-*-*}) GAS will not bump

past sparclite by default, an option must be passed to enable the

v9 instructions.

GAS treats sparclite as being compatible with v8, unless an architecture

is explicitly requested.  SPARC v9 is always incompatible with sparclite.

@c The order here is the same as the order of enum sparc_opcode_arch_val

@c to give the user a sense of the order of the "bumping".

@table @code

@kindex -Av6

@kindex Av7

@kindex -Av8

@kindex -Asparclet

@kindex -Asparclite

@kindex -Av9

@kindex -Av9a

@item -Av6 | -Av7 | -Av8 | -Asparclet | -Asparclite

@itemx -Av8plus | -Av8plusa | -Av9 | -Av9a

Use one of the @samp{-A} options to select one of the SPARC

architectures explicitly.  If you select an architecture explicitly,

@code{@value{AS}} reports a fatal error if it encounters an instruction

or feature requiring an incompatible or higher level.

@samp{-Av8plus} and @samp{-Av8plusa} select a 32 bit environment.

@samp{-Av9} and @samp{-Av9a} select a 64 bit environment and are not

available unless GAS is explicitly configured with 64 bit environment

support.

@samp{-Av8plusa} and @samp{-Av9a} enable the SPARC V9 instruction set with

UltraSPARC extensions.

@item -xarch=v8plus | -xarch=v8plusa

For compatibility with the SunOS v9 assembler.  These options are

equivalent to -Av8plus and -Av8plusa, respectively.

@item -bump

Warn whenever it is necessary to switch to another level.

If an architecture level is explicitly requested, GAS will not issue

warnings until that level is reached, and will then bump the level

as required (except between incompatible levels).

@item -32 | -64

Select the word size, either 32 bits or 64 bits.

These options are only available with the ELF object file format,

and require that the necessary BFD support has been included.

@end table

@node Sparc-Aligned-Data

@section Enforcing aligned data

@cindex data alignment on SPARC

@cindex SPARC data alignment

SPARC GAS normally permits data to be misaligned.  For example, it

permits the @code{.long} pseudo-op to be used on a byte boundary.

However, the native SunOS assemblers issue an error when they see

misaligned data.

@kindex --enforce-aligned-data

You can use the @code{--enforce-aligned-data} option to make SPARC GAS

also issue an error about misaligned data, just as the SunOS

assemblers do.

The @code{--enforce-aligned-data} option is not the default because gcc

issues misaligned data pseudo-ops when it initializes certain packed

data structures (structures defined using the @code{packed} attribute).

You may have to assemble with GAS in order to initialize packed data

structures in your own code.

@cindex SPARC syntax

@cindex syntax, SPARC

@node Sparc-Syntax

@section Sparc Syntax

The assembler syntax closely follows The Sparc Architecture Manual,

versions 8 and 9, as well as most extensions defined by Sun

for their UltraSPARC and Niagara line of processors.

@menu

* Sparc-Chars::                Special Characters

* Sparc-Regs::                 Register Names

* Sparc-Constants::            Constant Names

* Sparc-Relocs::               Relocations

* Sparc-Size-Translations::    Size Translations

@end menu

@node Sparc-Chars

@subsection Special Characters

@cindex line comment character, Sparc

@cindex Sparc line comment character

@samp{#} is the line comment character.

@cindex line separator, Sparc

@cindex statement separator, Sparc

@cindex Sparc line separator

@samp{;} can be used instead of a newline to separate statements.

@node Sparc-Regs

@subsection Register Names

@cindex Sparc registers

@cindex register names, Sparc

The Sparc integer register file is broken down into global,

outgoing, local, and incoming.

@itemize @bullet

@item

The 8 global registers are referred to as @samp{%g@var{n}}.

@item

The 8 outgoing registers are referred to as @samp{%o@var{n}}.

@item

The 8 local registers are referred to as @samp{%l@var{n}}.

@item

The 8 incoming registers are referred to as @samp{%i@var{n}}.

@item

The frame pointer register @samp{%i6} can be referenced using

the alias @samp{%fp}.

@item

The stack pointer register @samp{%o6} can be referenced using

the alias @samp{%sp}.

@end itemize

Floating point registers are simply referred to as @samp{%f@var{n}}.

When assembling for pre-V9, only 32 floating point registers

are available.  For V9 and later there are 64, but there are

restrictions when referencing the upper 32 registers.  They

can only be accessed as double or quad, and thus only even

or quad numbered accesses are allowed.  For example, @samp{%f34}

is a legal floating point register, but @samp{%f35} is not.

Certain V9 instructions allow access to ancillary state registers.

Most simply they can be referred to as @samp{%asr@var{n}} where

@var{n} can be from 16 to 31.  However, there are some aliases

defined to reference ASR registers defined for various UltraSPARC

processors:

@itemize @bullet

@item

The tick compare register is referred to as @samp{%tick_cmpr}.

@item

The system tick register is referred to as @samp{%stick}.  An alias,

@samp{%sys_tick}, exists but is deprecated and should not be used

by new software.

@item

The system tick compare register is referred to as @samp{%stick_cmpr}.

An alias, @samp{%sys_tick_cmpr}, exists but is deprecated and should

not be used by new software.

@item

The software interrupt register is referred to as @samp{%softint}.

@item

The set software interrupt register is referred to as @samp{%set_softint}.

The mnemonic @samp{%softint_set} is provided as an alias.

@item

The clear software interrupt register is referred to as

@samp{%clear_softint}.  The mnemonic @samp{%softint_clear} is provided

as an alias.

@item

The performance instrumentation counters register is referred to as

@samp{%pic}.

@item

The performance control register is referred to as @samp{%pcr}.

@item

The graphics status register is referred to as @samp{%gsr}.

@item

The V9 dispatch control register is referred to as @samp{%dcr}.

@end itemize

Various V9 branch and conditional move instructions allow

specification of which set of integer condition codes to

test.  These are referred to as @samp{%xcc} and @samp{%icc}.

In V9, there are 4 sets of floating point condition codes

which are referred to as @samp{%fcc@var{n}}.

Several special privileged and non-privileged registers

exist:

@itemize @bullet

@item

The V9 address space identifier register is referred to as @samp{%asi}.

@item

The V9 restorable windows register is referred to as @samp{%canrestore}.

@item

The V9 savable windows register is referred to as @samp{%cansave}.

@item

The V9 clean windows register is referred to as @samp{%cleanwin}.

@item

The V9 current window pointer register is referred to as @samp{%cwp}.

@item

The floating-point queue register is referred to as @samp{%fq}.

@item

The V8 co-processor queue register is referred to as @samp{%cq}.

@item

The floating point status register is referred to as @samp{%fsr}.

@item

The other windows register is referred to as @samp{%otherwin}.

@item

The V9 program counter register is referred to as @samp{%pc}.

@item

The V9 next program counter register is referred to as @samp{%npc}.

@item

The V9 processor interrupt level register is referred to as @samp{%pil}.

@item

The V9 processor state register is referred to as @samp{%pstate}.

@item

The trap base address register is referred to as @samp{%tba}.

@item

The V9 tick register is referred to as @samp{%tick}.

@item

The V9 trap level is referred to as @samp{%tl}.

@item

The V9 trap program counter is referred to as @samp{%tpc}.

@item

The V9 trap next program counter is referred to as @samp{%tnpc}.

@item

The V9 trap state is referred to as @samp{%tstate}.

@item

The V9 trap type is referred to as @samp{%tt}.

@item

The V9 condition codes is referred to as @samp{%ccr}.

@item

The V9 floating-point registers state is referred to as @samp{%fprs}.

@item

The V9 version register is referred to as @samp{%ver}.

@item

The V9 window state register is referred to as @samp{%wstate}.

@item

The Y register is referred to as @samp{%y}.

@item

The V8 window invalid mask register is referred to as @samp{%wim}.

@item

The V8 processor state register is referred to as @samp{%psr}.

@item

The V9 global register level register is referred to as @samp{%gl}.

@end itemize

Several special register names exist for hypervisor mode code:

@itemize @bullet

@item

The hyperprivileged processor state register is referred to as

@samp{%hpstate}.

@item

The hyperprivileged trap state register is referred to as @samp{%htstate}.

@item

The hyperprivileged interrupt pending register is referred to as

@samp{%hintp}.

@item

The hyperprivileged trap base address register is referred to as

@samp{%htba}.

@item

The hyperprivileged implementation version register is referred

to as @samp{%hver}.

@item

The hyperprivileged system tick compare register is referred

to as @samp{%hstick_cmpr}.  Note that there is no @samp{%hstick}

register, the normal @samp{%stick} is used.

@end itemize

@node Sparc-Constants

@subsection Constants

@cindex Sparc constants

@cindex constants, Sparc

Several Sparc instructions take an immediate operand field for

which mnemonic names exist.  Two such examples are @samp{membar}

and @samp{prefetch}.  Another example are the set of V9

memory access instruction that allow specification of an

address space identifier.

The @samp{membar} instruction specifies a memory barrier that is

the defined by the operand which is a bitmask.  The supported

mask mnemonics are:

@itemize @bullet

@item

@samp{#Sync} requests that all operations (including nonmemory

reference operations) appearing prior to the @code{membar} must have

been performed and the effects of any exceptions become visible before

any instructions after the @code{membar} may be initiated.  This

corresponds to @code{membar} cmask field bit 2.

@item

@samp{#MemIssue} requests that all memory reference operations

appearing prior to the @code{membar} must have been performed before

any memory operation after the @code{membar} may be initiated.  This

corresponds to @code{membar} cmask field bit 1.

@item

@samp{#Lookaside} requests that a store appearing prior to the

@code{membar} must complete before any load following the

@code{membar} referencing the same address can be initiated.  This

corresponds to @code{membar} cmask field bit 0.

@item

@samp{#StoreStore} defines that the effects of all stores appearing

prior to the @code{membar} instruction must be visible to all

processors before the effect of any stores following the

@code{membar}.  Equivalent to the deprecated @code{stbar} instruction.

This corresponds to @code{membar} mmask field bit 3.

@item

@samp{#LoadStore} defines all loads appearing prior to the

@code{membar} instruction must have been performed before the effect

of any stores following the @code{membar} is visible to any other

processor.  This corresponds to @code{membar} mmask field bit 2.

@item

@samp{#StoreLoad} defines that the effects of all stores appearing

prior to the @code{membar} instruction must be visible to all

processors before loads following the @code{membar} may be performed.

This corresponds to @code{membar} mmask field bit 1.

@item

@samp{#LoadLoad} defines that all loads appearing prior to the

@code{membar} instruction must have been performed before any loads

following the @code{membar} may be performed.  This corresponds to

@code{membar} mmask field bit 0.

@end itemize

These values can be ored together, for example:

@example

membar #Sync

membar #StoreLoad | #LoadLoad

membar #StoreLoad | #StoreStore

@end example

The @code{prefetch} and @code{prefetcha} instructions take a prefetch

function code.  The following prefetch function code constant

mnemonics are available:

@itemize @bullet

@item

@samp{#n_reads} requests a prefetch for several reads, and corresponds

to a prefetch function code of 0.

@samp{#one_read} requests a prefetch for one read, and corresponds

to a prefetch function code of 1.

@samp{#n_writes} requests a prefetch for several writes (and possibly

reads), and corresponds to a prefetch function code of 2.

@samp{#one_write} requests a prefetch for one write, and corresponds

to a prefetch function code of 3.

@samp{#page} requests a prefetch page, and corresponds to a prefetch

function code of 4.

@samp{#invalidate} requests a prefetch invalidate, and corresponds to

a prefetch function code of 16.

@samp{#unified} requests a prefetch to the nearest unified cache, and

corresponds to a prefetch function code of 17.

@samp{#n_reads_strong} requests a strong prefetch for several reads,

and corresponds to a prefetch function code of 20.

@samp{#one_read_strong} requests a strong prefetch for one read,

and corresponds to a prefetch function code of 21.

@samp{#n_writes_strong} requests a strong prefetch for several writes,

and corresponds to a prefetch function code of 22.

@samp{#one_write_strong} requests a strong prefetch for one write,

and corresponds to a prefetch function code of 23.

Onle one prefetch code may be specified.  Here are some examples:

@example

prefetch  [%l0 + %l2], #one_read

prefetch  [%g2 + 8], #n_writes

prefetcha [%g1] 0x8, #unified

prefetcha [%o0 + 0x10] %asi, #n_reads

@end example

The actual behavior of a given prefetch function code is processor

specific.  If a processor does not implement a given prefetch

function code, it will treat the prefetch instruction as a nop.

For instructions that accept an immediate address space identifier,

@code{@value{AS}} provides many mnemonics corresponding to

V9 defined as well as UltraSPARC and Niagara extended values.

For example, @samp{#ASI_P} and @samp{#ASI_BLK_INIT_QUAD_LDD_AIUS}.

See the V9 and processor specific manuals for details.

@end itemize

@node Sparc-Relocs

@subsection Relocations

@cindex Sparc relocations

@cindex relocations, Sparc

ELF relocations are available as defined in the 32-bit and 64-bit

Sparc ELF specifications.

@code{R_SPARC_HI22} is obtained using @samp{%hi} and @code{R_SPARC_LO10}

is obtained using @samp{%lo}.  Likewise @code{R_SPARC_HIX22} is

obtained from @samp{%hix} and @code{R_SPARC_LOX10} is obtained

using @samp{%lox}.  For example:

@example

sethi %hi(symbol), %g1

or    %g1, %lo(symbol), %g1

sethi %hix(symbol), %g1

xor   %g1, %lox(symbol), %g1

@end example

These ``high'' mnemonics extract bits 31:10 of their operand,

and the ``low'' mnemonics extract bits 9:0 of their operand.

V9 code model relocations can be requested as follows:

@itemize @bullet

@item

@code{R_SPARC_HH22} is requested using @samp{%hh}.  It can

also be generated using @samp{%uhi}.

@item

@code{R_SPARC_HM10} is requested using @samp{%hm}.  It can

also be generated using @samp{%ulo}.

@item

@code{R_SPARC_LM22} is requested using @samp{%lm}.

@item

@code{R_SPARC_H44} is requested using @samp{%h44}.

@item

@code{R_SPARC_M44} is requested using @samp{%m44}.

@item

@code{R_SPARC_L44} is requested using @samp{%l44}.

@end itemize

The PC relative relocation @code{R_SPARC_PC22} can be obtained by

enclosing an operand inside of @samp{%pc22}.  Likewise, the

@code{R_SPARC_PC10} relocation can be obtained using @samp{%pc10}.

These are mostly used when assembling PIC code.  For example, the

standard PIC sequence on Sparc to get the base of the global offset

table, PC relative, into a register, can be performed as:

@example

sethi %pc22(_GLOBAL_OFFSET_TABLE_-4), %l7

add   %l7, %pc10(_GLOBAL_OFFSET_TABLE_+4), %l7

@end example

Several relocations exist to allow the link editor to potentially

optimize GOT data references.  The @code{R_SPARC_GOTDATA_OP_HIX22}

relocation can obtained by enclosing an operand inside of

@samp{%gdop_hix22}.  The @code{R_SPARC_GOTDATA_OP_LOX10}

relocation can obtained by enclosing an operand inside of

@samp{%gdop_lox10}.  Likewise, @code{R_SPARC_GOTDATA_OP} can be

obtained by enclosing an operand inside of @samp{%gdop}.

For example, assuming the GOT base is in register @code{%l7}:

@example

sethi %gdop_hix22(symbol), %l1

xor   %l1, %gdop_lox10(symbol), %l1

ld    [%l7 + %l1], %l2, %gdop(symbol)

@end example

There are many relocations that can be requested for access to

thread local storage variables.  All of the Sparc TLS mnemonics

are supported:

@itemize @bullet

@item

@code{R_SPARC_TLS_GD_HI22} is requested using @samp{%tgd_hi22}.

@item

@code{R_SPARC_TLS_GD_LO10} is requested using @samp{%tgd_lo10}.

@item

@code{R_SPARC_TLS_GD_ADD} is requested using @samp{%tgd_add}.

@item

@code{R_SPARC_TLS_GD_CALL} is requested using @samp{%tgd_call}.

@item

@code{R_SPARC_TLS_LDM_HI22} is requested using @samp{%tldm_hi22}.

@item

@code{R_SPARC_TLS_LDM_LO10} is requested using @samp{%tldm_lo10}.

@item

@code{R_SPARC_TLS_LDM_ADD} is requested using @samp{%tldm_add}.

@item

@code{R_SPARC_TLS_LDM_CALL} is requested using @samp{%tldm_call}.

@item

@code{R_SPARC_TLS_LDO_HIX22} is requested using @samp{%tldo_hix22}.

@item

@code{R_SPARC_TLS_LDO_LOX10} is requested using @samp{%tldo_lox10}.

@item

@code{R_SPARC_TLS_LDO_ADD} is requested using @samp{%tldo_add}.

@item

@code{R_SPARC_TLS_IE_HI22} is requested using @samp{%tie_hi22}.

@item

@code{R_SPARC_TLS_IE_LO10} is requested using @samp{%tie_lo10}.

@item

@code{R_SPARC_TLS_IE_LD} is requested using @samp{%tie_ld}.

@item

@code{R_SPARC_TLS_IE_LDX} is requested using @samp{%tie_ldx}.

@item

@code{R_SPARC_TLS_IE_ADD} is requested using @samp{%tie_add}.

@item

@code{R_SPARC_TLS_LE_HIX22} is requested using @samp{%tle_hix22}.

@item

@code{R_SPARC_TLS_LE_LOX10} is requested using @samp{%tle_lox10}.

@end itemize

Here are some example TLS model sequences.

First, General Dynamic:

@example

sethi  %tgd_hi22(symbol), %l1

add    %l1, %tgd_lo10(symbol), %l1

add    %l7, %l1, %o0, %tgd_add(symbol)

call   __tls_get_addr, %tgd_call(symbol)

nop

@end example

Local Dynamic:

@example

sethi  %tldm_hi22(symbol), %l1

add    %l1, %tldm_lo10(symbol), %l1

add    %l7, %l1, %o0, %tldm_add(symbol)

call   __tls_get_addr, %tldm_call(symbol)

nop

sethi  %tldo_hix22(symbol), %l1

xor    %l1, %tldo_lox10(symbol), %l1

add    %o0, %l1, %l1, %tldo_add(symbol)

@end example

Initial Exec:

@example

sethi  %tie_hi22(symbol), %l1

add    %l1, %tie_lo10(symbol), %l1

ld     [%l7 + %l1], %o0, %tie_ld(symbol)

add    %g7, %o0, %o0, %tie_add(symbol)

sethi  %tie_hi22(symbol), %l1

add    %l1, %tie_lo10(symbol), %l1

ldx    [%l7 + %l1], %o0, %tie_ldx(symbol)

add    %g7, %o0, %o0, %tie_add(symbol)

@end example

And finally, Local Exec:

@example

sethi  %tle_hix22(symbol), %l1

add    %l1, %tle_lox10(symbol), %l1

add    %g7, %l1, %l1

@end example

When assembling for 64-bit, and a secondary constant addend is

specified in an address expression that would normally generate

an @code{R_SPARC_LO10} relocation, the assembler will emit an

@code{R_SPARC_OLO10} instead.

@node Sparc-Size-Translations

@subsection Size Translations

@cindex Sparc size translations

@cindex size, translations, Sparc

Often it is desirable to write code in an operand size agnostic

manner.  @code{@value{AS}} provides support for this via

operand size opcode translations.  Translations are supported

for loads, stores, shifts, compare-and-swap atomics, and the

@samp{clr} synthetic instruction.

If generating 32-bit code, @code{@value{AS}} will generate the

32-bit opcode.  Whereas if 64-bit code is being generated,

the 64-bit opcode will be emitted.  For example @code{ldn}

will be transformed into @code{ld} for 32-bit code and

@code{ldx} for 64-bit code.

Here is an example meant to demonstrate all the supported

opcode translations:

@example

ldn   [%o0], %o1

ldna  [%o0] %asi, %o2

stn   %o1, [%o0]

stna  %o2, [%o0] %asi

slln  %o3, 3, %o3

srln  %o4, 8, %o4

sran  %o5, 12, %o5

casn  [%o0], %o1, %o2

casna [%o0] %asi, %o1, %o2

clrn  %g1

@end example

In 32-bit mode @code{@value{AS}} will emit:

@example

ld   [%o0], %o1

lda  [%o0] %asi, %o2

st   %o1, [%o0]

sta  %o2, [%o0] %asi

sll  %o3, 3, %o3

srl  %o4, 8, %o4

sra  %o5, 12, %o5

cas  [%o0], %o1, %o2

casa [%o0] %asi, %o1, %o2

clr  %g1

@end example

And in 64-bit mode @code{@value{AS}} will emit:

@example

ldx   [%o0], %o1

ldxa  [%o0] %asi, %o2

stx   %o1, [%o0]

stxa  %o2, [%o0] %asi

sllx  %o3, 3, %o3

srlx  %o4, 8, %o4

srax  %o5, 12, %o5

casx  [%o0], %o1, %o2

casxa [%o0] %asi, %o1, %o2

clrx  %g1

@end example

Finally, the @samp{.nword} translating directive is supported

as well.  It is documented in the section on Sparc machine

directives.

@node Sparc-Float

@section Floating Point

@cindex floating point, SPARC (@sc{ieee})

@cindex SPARC floating point (@sc{ieee})

The Sparc uses @sc{ieee} floating-point numbers.

@node Sparc-Directives

@section Sparc Machine Directives

@cindex SPARC machine directives

@cindex machine directives, SPARC

The Sparc version of @code{@value{AS}} supports the following additional

machine directives:

@table @code

@cindex @code{align} directive, SPARC

@item .align

This must be followed by the desired alignment in bytes.

@cindex @code{common} directive, SPARC

@item .common

This must be followed by a symbol name, a positive number, and

@code{"bss"}.  This behaves somewhat like @code{.comm}, but the

syntax is different.

@cindex @code{half} directive, SPARC

@item .half

This is functionally identical to @code{.short}.

@cindex @code{nword} directive, SPARC

@item .nword

On the Sparc, the @code{.nword} directive produces native word sized value,

ie. if assembling with -32 it is equivalent to @code{.word}, if assembling

with -64 it is equivalent to @code{.xword}.

@cindex @code{proc} directive, SPARC

@item .proc

This directive is ignored.  Any text following it on the same

line is also ignored.

@cindex @code{register} directive, SPARC

@item .register

This directive declares use of a global application or system register.

It must be followed by a register name %g2, %g3, %g6 or %g7, comma and

the symbol name for that register.  If symbol name is @code{#scratch},

it is a scratch register, if it is @code{#ignore}, it just suppresses any

errors about using undeclared global register, but does not emit any

information about it into the object file.  This can be useful e.g. if you

save the register before use and restore it after.

@cindex @code{reserve} directive, SPARC

@item .reserve

This must be followed by a symbol name, a positive number, and

@code{"bss"}.  This behaves somewhat like @code{.lcomm}, but the

syntax is different.

@cindex @code{seg} directive, SPARC

@item .seg

This must be followed by @code{"text"}, @code{"data"}, or

@code{"data1"}.  It behaves like @code{.text}, @code{.data}, or

@code{.data 1}.

@cindex @code{skip} directive, SPARC

@item .skip

This is functionally identical to the @code{.space} directive.

@cindex @code{word} directive, SPARC

@item .word

On the Sparc, the @code{.word} directive produces 32 bit values,

instead of the 16 bit values it produces on many other machines.

@cindex @code{xword} directive, SPARC

@item .xword

On the Sparc V9 processor, the @code{.xword} directive produces

64 bit values.

@end table