gas/doc/c-xtensa.texi

@c Copyright (C) 2002-2021 Free Software Foundation, Inc.
@c This is part of the GAS manual.
@c For copying conditions, see the file as.texinfo.
@c
@c man end
@ifset GENERIC
@page
@node Xtensa-Dependent
@chapter Xtensa Dependent Features
@end ifset
@ifclear GENERIC
@node Machine Dependencies
@chapter Xtensa Dependent Features
@end ifclear

@cindex Xtensa architecture
This chapter covers features of the @sc{gnu} assembler that are specific
to the Xtensa architecture.  For details about the Xtensa instruction
set, please consult the @cite{Xtensa Instruction Set Architecture (ISA)
Reference Manual}.

@menu
* Xtensa Options::              Command-line Options.
* Xtensa Syntax::               Assembler Syntax for Xtensa Processors.
* Xtensa Optimizations::        Assembler Optimizations.
* Xtensa Relaxation::           Other Automatic Transformations.
* Xtensa Directives::           Directives for Xtensa Processors.
@end menu

@node Xtensa Options
@section Command-line Options

@c man begin OPTIONS
@table @gcctabopt

@item --text-section-literals | --no-text-section-literals
@kindex --text-section-literals
@kindex --no-text-section-literals
Control the treatment of literal pools.  The default is
@samp{--no-@-text-@-section-@-literals}, which places literals in
separate sections in the output file.  This allows the literal pool to be
placed in a data RAM/ROM.  With @samp{--text-@-section-@-literals}, the
literals are interspersed in the text section in order to keep them as
close as possible to their references.  This may be necessary for large
assembly files, where the literals would otherwise be out of range of the
@code{L32R} instructions in the text section.  Literals are grouped into
pools following @code{.literal_position} directives or preceding
@code{ENTRY} instructions.  These options only affect literals referenced
via PC-relative @code{L32R} instructions; literals for absolute mode
@code{L32R} instructions are handled separately.
@xref{Literal Directive, ,literal}.

@item --auto-litpools | --no-auto-litpools
@kindex --auto-litpools
@kindex --no-auto-litpools
Control the treatment of literal pools.  The default is
@samp{--no-@-auto-@-litpools}, which in the absence of
@samp{--text-@-section-@-literals} places literals in separate sections
in the output file.  This allows the literal pool to be placed in a data
RAM/ROM.  With @samp{--auto-@-litpools}, the literals are interspersed
in the text section in order to keep them as close as possible to their
references, explicit @code{.literal_position} directives are not
required.  This may be necessary for very large functions, where single
literal pool at the beginning of the function may not be reachable by
@code{L32R} instructions at the end.  These options only affect
literals referenced via PC-relative @code{L32R} instructions; literals
for absolute mode @code{L32R} instructions are handled separately.
When used together with @samp{--text-@-section-@-literals},
@samp{--auto-@-litpools} takes precedence.
@xref{Literal Directive, ,literal}.

@item --absolute-literals | --no-absolute-literals
@kindex --absolute-literals
@kindex --no-absolute-literals
Indicate to the assembler whether @code{L32R} instructions use absolute
or PC-relative addressing.  If the processor includes the absolute
addressing option, the default is to use absolute @code{L32R}
relocations.  Otherwise, only the PC-relative @code{L32R} relocations
can be used.

@item --target-align | --no-target-align
@kindex --target-align
@kindex --no-target-align
Enable or disable automatic alignment to reduce branch penalties at some
expense in code size.  @xref{Xtensa Automatic Alignment, ,Automatic
Instruction Alignment}.  This optimization is enabled by default.  Note
that the assembler will always align instructions like @code{LOOP} that
have fixed alignment requirements.

@item --longcalls | --no-longcalls
@kindex --longcalls
@kindex --no-longcalls
Enable or disable transformation of call instructions to allow calls
across a greater range of addresses.  @xref{Xtensa Call Relaxation,
,Function Call Relaxation}.  This option should be used when call
targets can potentially be out of range.  It may degrade both code size
and performance, but the linker can generally optimize away the
unnecessary overhead when a call ends up within range.  The default is
@samp{--no-@-longcalls}.

@item --transform | --no-transform
@kindex --transform
@kindex --no-transform
Enable or disable all assembler transformations of Xtensa instructions,
including both relaxation and optimization.  The default is
@samp{--transform}; @samp{--no-transform} should only be used in the
rare cases when the instructions must be exactly as specified in the
assembly source.  Using @samp{--no-transform} causes out of range
instruction operands to be errors.

@item --rename-section @var{oldname}=@var{newname}
@kindex --rename-section
Rename the @var{oldname} section to @var{newname}.  This option can be used
multiple times to rename multiple sections.

@item --trampolines | --no-trampolines
@kindex --trampolines
@kindex --no-trampolines
Enable or disable transformation of jump instructions to allow jumps
across a greater range of addresses.  @xref{Xtensa Jump Relaxation,
,Jump Trampolines}.  This option should be used when jump targets can
potentially be out of range.  In the absence of such jumps this option
does not affect code size or performance.  The default is
@samp{--trampolines}.

@item --abi-windowed | --abi-call0
@kindex --abi-windowed
@kindex --abi-call0
Choose ABI tag written to the @code{.xtensa.info} section.  ABI tag
indicates ABI of the assembly code.  A warning is issued by the linker
on an attempt to link object files with inconsistent ABI tags.
Default ABI is chosen by the Xtensa core configuration.
@end table

@c man end

@node Xtensa Syntax
@section Assembler Syntax
@cindex syntax, Xtensa assembler
@cindex Xtensa assembler syntax
@cindex FLIX syntax

Block comments are delimited by @samp{/*} and @samp{*/}.  End of line
comments may be introduced with either @samp{#} or @samp{//}.

If a @samp{#} appears as the first character of a line then the whole
line is treated as a comment, but in this case the line could also be
a logical line number directive (@pxref{Comments}) or a preprocessor
control command (@pxref{Preprocessing}).

Instructions consist of a leading opcode or macro name followed by
whitespace and an optional comma-separated list of operands:

@smallexample
@var{opcode} [@var{operand}, @dots{}]
@end smallexample

Instructions must be separated by a newline or semicolon (@samp{;}).

FLIX instructions, which bundle multiple opcodes together in a single
instruction, are specified by enclosing the bundled opcodes inside
braces:

@smallexample
@group
@{
[@var{format}]
@var{opcode0} [@var{operands}]
@end group
@var{opcode1} [@var{operands}]
@group
@var{opcode2} [@var{operands}]
@dots{}
@}
@end group
@end smallexample

The opcodes in a FLIX instruction are listed in the same order as the
corresponding instruction slots in the TIE format declaration.
Directives and labels are not allowed inside the braces of a FLIX
instruction.  A particular TIE format name can optionally be specified
immediately after the opening brace, but this is usually unnecessary.
The assembler will automatically search for a format that can encode the
specified opcodes, so the format name need only be specified in rare
cases where there is more than one applicable format and where it
matters which of those formats is used.  A FLIX instruction can also be
specified on a single line by separating the opcodes with semicolons:

@smallexample
@{ [@var{format};] @var{opcode0} [@var{operands}]; @var{opcode1} [@var{operands}]; @var{opcode2} [@var{operands}]; @dots{} @}
@end smallexample

If an opcode can only be encoded in a FLIX instruction but is not
specified as part of a FLIX bundle, the assembler will choose the
smallest format where the opcode can be encoded and
will fill unused instruction slots with no-ops.

@menu
* Xtensa Opcodes::              Opcode Naming Conventions.
* Xtensa Registers::            Register Naming.
@end menu

@node Xtensa Opcodes
@subsection Opcode Names
@cindex Xtensa opcode names
@cindex opcode names, Xtensa

See the @cite{Xtensa Instruction Set Architecture (ISA) Reference
Manual} for a complete list of opcodes and descriptions of their
semantics.

@cindex _ opcode prefix
If an opcode name is prefixed with an underscore character (@samp{_}),
@command{@value{AS}} will not transform that instruction in any way.  The
underscore prefix disables both optimization (@pxref{Xtensa
Optimizations, ,Xtensa Optimizations}) and relaxation (@pxref{Xtensa
Relaxation, ,Xtensa Relaxation}) for that particular instruction.  Only
use the underscore prefix when it is essential to select the exact
opcode produced by the assembler.  Using this feature unnecessarily
makes the code less efficient by disabling assembler optimization and
less flexible by disabling relaxation.

Note that this special handling of underscore prefixes only applies to
Xtensa opcodes, not to either built-in macros or user-defined macros.
When an underscore prefix is used with a macro (e.g., @code{_MOV}), it
refers to a different macro.  The assembler generally provides built-in
macros both with and without the underscore prefix, where the underscore
versions behave as if the underscore carries through to the instructions
in the macros.  For example, @code{_MOV} may expand to @code{_MOV.N}@.

The underscore prefix only applies to individual instructions, not to
series of instructions.  For example, if a series of instructions have
underscore prefixes, the assembler will not transform the individual
instructions, but it may insert other instructions between them (e.g.,
to align a @code{LOOP} instruction).  To prevent the assembler from
modifying a series of instructions as a whole, use the
@code{no-transform} directive.  @xref{Transform Directive, ,transform}.

@node Xtensa Registers
@subsection Register Names
@cindex Xtensa register names
@cindex register names, Xtensa
@cindex sp register

The assembly syntax for a register file entry is the ``short'' name for
a TIE register file followed by the index into that register file.  For
example, the general-purpose @code{AR} register file has a short name of
@code{a}, so these registers are named @code{a0}@dots{}@code{a15}.
As a special feature, @code{sp} is also supported as a synonym for
@code{a1}.  Additional registers may be added by processor configuration
options and by designer-defined TIE extensions.  An initial @samp{$}
character is optional in all register names.

@node Xtensa Optimizations
@section Xtensa Optimizations
@cindex optimizations

The optimizations currently supported by @command{@value{AS}} are
generation of density instructions where appropriate and automatic
branch target alignment.

@menu
* Density Instructions::        Using Density Instructions.
* Xtensa Automatic Alignment::  Automatic Instruction Alignment.
@end menu

@node Density Instructions
@subsection Using Density Instructions
@cindex density instructions

The Xtensa instruction set has a code density option that provides
16-bit versions of some of the most commonly used opcodes.  Use of these
opcodes can significantly reduce code size.  When possible, the
assembler automatically translates instructions from the core
Xtensa instruction set into equivalent instructions from the Xtensa code
density option.  This translation can be disabled by using underscore
prefixes (@pxref{Xtensa Opcodes, ,Opcode Names}), by using the
@samp{--no-transform} command-line option (@pxref{Xtensa Options, ,Command
Line Options}), or by using the @code{no-transform} directive
(@pxref{Transform Directive, ,transform}).

It is a good idea @emph{not} to use the density instructions directly.
The assembler will automatically select dense instructions where
possible.  If you later need to use an Xtensa processor without the code
density option, the same assembly code will then work without modification.

@node Xtensa Automatic Alignment
@subsection Automatic Instruction Alignment
@cindex alignment of @code{LOOP} instructions
@cindex alignment of branch targets
@cindex @code{LOOP} instructions, alignment
@cindex branch target alignment

The Xtensa assembler will automatically align certain instructions, both
to optimize performance and to satisfy architectural requirements.

As an optimization to improve performance, the assembler attempts to
align branch targets so they do not cross instruction fetch boundaries.
(Xtensa processors can be configured with either 32-bit or 64-bit
instruction fetch widths.)  An
instruction immediately following a call is treated as a branch target
in this context, because it will be the target of a return from the
call.  This alignment has the potential to reduce branch penalties at
some expense in code size.
This optimization is enabled by default.  You can disable it with the
@samp{--no-target-@-align} command-line option (@pxref{Xtensa Options,
,Command-line Options}).

The target alignment optimization is done without adding instructions
that could increase the execution time of the program.  If there are
density instructions in the code preceding a target, the assembler can
change the target alignment by widening some of those instructions to
the equivalent 24-bit instructions.  Extra bytes of padding can be
inserted immediately following unconditional jump and return
instructions.
This approach is usually successful in aligning many, but not all,
branch targets.

The @code{LOOP} family of instructions must be aligned such that the
first instruction in the loop body does not cross an instruction fetch
boundary (e.g., with a 32-bit fetch width, a @code{LOOP} instruction
must be on either a 1 or 2 mod 4 byte boundary).  The assembler knows
about this restriction and inserts the minimal number of 2 or 3 byte
no-op instructions to satisfy it.  When no-op instructions are added,
any label immediately preceding the original loop will be moved in order
to refer to the loop instruction, not the newly generated no-op
instruction.  To preserve binary compatibility across processors with
different fetch widths, the assembler conservatively assumes a 32-bit
fetch width when aligning @code{LOOP} instructions (except if the first
instruction in the loop is a 64-bit instruction).

Previous versions of the assembler automatically aligned @code{ENTRY}
instructions to 4-byte boundaries, but that alignment is now the
programmer's responsibility.

@node Xtensa Relaxation
@section Xtensa Relaxation
@cindex relaxation

When an instruction operand is outside the range allowed for that
particular instruction field, @command{@value{AS}} can transform the code
to use a functionally-equivalent instruction or sequence of
instructions.  This process is known as @dfn{relaxation}.  This is
typically done for branch instructions because the distance of the
branch targets is not known until assembly-time.  The Xtensa assembler
offers branch relaxation and also extends this concept to function
calls, @code{MOVI} instructions and other instructions with immediate
fields.

@menu
* Xtensa Branch Relaxation::        Relaxation of Branches.
* Xtensa Call Relaxation::          Relaxation of Function Calls.
* Xtensa Jump Relaxation::          Relaxation of Jumps.
* Xtensa Immediate Relaxation::     Relaxation of other Immediate Fields.
@end menu

@node Xtensa Branch Relaxation
@subsection Conditional Branch Relaxation
@cindex relaxation of branch instructions
@cindex branch instructions, relaxation

When the target of a branch is too far away from the branch itself,
i.e., when the offset from the branch to the target is too large to fit
in the immediate field of the branch instruction, it may be necessary to
replace the branch with a branch around a jump.  For example,

@smallexample
    beqz    a2, L
@end smallexample

may result in:

@smallexample
@group
    bnez.n  a2, M
    j L
M:
@end group
@end smallexample

(The @code{BNEZ.N} instruction would be used in this example only if the
density option is available.  Otherwise, @code{BNEZ} would be used.)

This relaxation works well because the unconditional jump instruction
has a much larger offset range than the various conditional branches.
However, an error will occur if a branch target is beyond the range of a
jump instruction.  @command{@value{AS}} cannot relax unconditional jumps.
Similarly, an error will occur if the original input contains an
unconditional jump to a target that is out of range.

Branch relaxation is enabled by default.  It can be disabled by using
underscore prefixes (@pxref{Xtensa Opcodes, ,Opcode Names}), the
@samp{--no-transform} command-line option (@pxref{Xtensa Options,
,Command-line Options}), or the @code{no-transform} directive
(@pxref{Transform Directive, ,transform}).

@node Xtensa Call Relaxation
@subsection Function Call Relaxation
@cindex relaxation of call instructions
@cindex call instructions, relaxation

Function calls may require relaxation because the Xtensa immediate call
instructions (@code{CALL0}, @code{CALL4}, @code{CALL8} and
@code{CALL12}) provide a PC-relative offset of only 512 Kbytes in either
direction.  For larger programs, it may be necessary to use indirect
calls (@code{CALLX0}, @code{CALLX4}, @code{CALLX8} and @code{CALLX12})
where the target address is specified in a register.  The Xtensa
assembler can automatically relax immediate call instructions into
indirect call instructions.  This relaxation is done by loading the
address of the called function into the callee's return address register
and then using a @code{CALLX} instruction.  So, for example:

@smallexample
    call8 func
@end smallexample

might be relaxed to:

@smallexample
@group
    .literal .L1, func
    l32r    a8, .L1
    callx8  a8
@end group
@end smallexample

Because the addresses of targets of function calls are not generally
known until link-time, the assembler must assume the worst and relax all
the calls to functions in other source files, not just those that really
will be out of range.  The linker can recognize calls that were
unnecessarily relaxed, and it will remove the overhead introduced by the
assembler for those cases where direct calls are sufficient.

Call relaxation is disabled by default because it can have a negative
effect on both code size and performance, although the linker can
usually eliminate the unnecessary overhead.  If a program is too large
and some of the calls are out of range, function call relaxation can be
enabled using the @samp{--longcalls} command-line option or the
@code{longcalls} directive (@pxref{Longcalls Directive, ,longcalls}).

@node Xtensa Jump Relaxation
@subsection Jump Relaxation
@cindex relaxation of jump instructions
@cindex jump instructions, relaxation

Jump instruction may require relaxation because the Xtensa jump instruction
(@code{J}) provide a PC-relative offset of only 128 Kbytes in either
direction.  One option is to use jump long (@code{J.L}) instruction, which
depending on jump distance may be assembled as jump (@code{J}) or indirect
jump (@code{JX}).  However it needs a free register.  When there's no spare
register it is possible to plant intermediate jump sites (trampolines)
between the jump instruction and its target.  These sites may be located in
areas unreachable by normal code execution flow, in that case they only
contain intermediate jumps, or they may be inserted in the middle of code
block, in which case there's an additional jump from the beginning of the
trampoline to the instruction past its end.  So, for example:

@smallexample
@group
    j 1f
    ...
    retw
    ...
    mov a10, a2
    call8 func
    ...
1:
    ...
@end group
@end smallexample

might be relaxed to:

@smallexample
@group
    j .L0_TR_1
    ...
    retw
.L0_TR_1:
    j 1f
    ...
    mov a10, a2
    call8 func
    ...
1:
    ...
@end group
@end smallexample

or to:

@smallexample
@group
    j .L0_TR_1
    ...
    retw
    ...
    mov a10, a2
    j .L0_TR_0
.L0_TR_1:
    j 1f
.L0_TR_0:
    call8 func
    ...
1:
    ...
@end group
@end smallexample

The Xtensa assembler uses trampolines with jump around only when it cannot
find suitable unreachable trampoline.  There may be multiple trampolines
between the jump instruction and its target.

This relaxation does not apply to jumps to undefined symbols, assuming they
will reach their targets once resolved.

Jump relaxation is enabled by default because it does not affect code size
or performance while the code itself is small.  This relaxation may be
disabled completely with @samp{--no-trampolines} or @samp{--no-transform}
command-line options (@pxref{Xtensa Options, ,Command-line Options}).

@node Xtensa Immediate Relaxation
@subsection Other Immediate Field Relaxation
@cindex immediate fields, relaxation
@cindex relaxation of immediate fields

The assembler normally performs the following other relaxations.  They
can be disabled by using underscore prefixes (@pxref{Xtensa Opcodes,
,Opcode Names}), the @samp{--no-transform} command-line option
(@pxref{Xtensa Options, ,Command-line Options}), or the
@code{no-transform} directive (@pxref{Transform Directive, ,transform}).

@cindex @code{MOVI} instructions, relaxation
@cindex relaxation of @code{MOVI} instructions
The @code{MOVI} machine instruction can only materialize values in the
range from -2048 to 2047.  Values outside this range are best
materialized with @code{L32R} instructions.  Thus:

@smallexample
    movi a0, 100000
@end smallexample

is assembled into the following machine code:

@smallexample
@group
    .literal .L1, 100000
    l32r a0, .L1
@end group
@end smallexample

@cindex @code{L8UI} instructions, relaxation
@cindex @code{L16SI} instructions, relaxation
@cindex @code{L16UI} instructions, relaxation
@cindex @code{L32I} instructions, relaxation
@cindex relaxation of @code{L8UI} instructions
@cindex relaxation of @code{L16SI} instructions
@cindex relaxation of @code{L16UI} instructions
@cindex relaxation of @code{L32I} instructions
The @code{L8UI} machine instruction can only be used with immediate
offsets in the range from 0 to 255. The @code{L16SI} and @code{L16UI}
machine instructions can only be used with offsets from 0 to 510.  The
@code{L32I} machine instruction can only be used with offsets from 0 to
1020.  A load offset outside these ranges can be materialized with
an @code{L32R} instruction if the destination register of the load
is different than the source address register.  For example:

@smallexample
    l32i a1, a0, 2040
@end smallexample

is translated to:

@smallexample
@group
    .literal .L1, 2040
    l32r a1, .L1
@end group
@group
    add a1, a0, a1
    l32i a1, a1, 0
@end group
@end smallexample

@noindent
If the load destination and source address register are the same, an
out-of-range offset causes an error.

@cindex @code{ADDI} instructions, relaxation
@cindex relaxation of @code{ADDI} instructions
The Xtensa @code{ADDI} instruction only allows immediate operands in the
range from -128 to 127.  There are a number of alternate instruction
sequences for the @code{ADDI} operation.  First, if the
immediate is 0, the @code{ADDI} will be turned into a @code{MOV.N}
instruction (or the equivalent @code{OR} instruction if the code density
option is not available).  If the @code{ADDI} immediate is outside of
the range -128 to 127, but inside the range -32896 to 32639, an
@code{ADDMI} instruction or @code{ADDMI}/@code{ADDI} sequence will be
used.  Finally, if the immediate is outside of this range and a free
register is available, an @code{L32R}/@code{ADD} sequence will be used
with a literal allocated from the literal pool.

For example:

@smallexample
@group
    addi    a5, a6, 0
    addi    a5, a6, 512
@end group
@group
    addi    a5, a6, 513
    addi    a5, a6, 50000
@end group
@end smallexample

is assembled into the following:

@smallexample
@group
    .literal .L1, 50000
    mov.n   a5, a6
@end group
    addmi   a5, a6, 0x200
    addmi   a5, a6, 0x200
    addi    a5, a5, 1
@group
    l32r    a5, .L1
    add     a5, a6, a5
@end group
@end smallexample

@node Xtensa Directives
@section Directives
@cindex Xtensa directives
@cindex directives, Xtensa

The Xtensa assembler supports a region-based directive syntax:

@smallexample
@group
    .begin @var{directive} [@var{options}]
    @dots{}
    .end @var{directive}
@end group
@end smallexample

All the Xtensa-specific directives that apply to a region of code use
this syntax.

The directive applies to code between the @code{.begin} and the
@code{.end}.  The state of the option after the @code{.end} reverts to
what it was before the @code{.begin}.
A nested @code{.begin}/@code{.end} region can further
change the state of the directive without having to be aware of its
outer state.  For example, consider:

@smallexample
@group
    .begin no-transform
L:  add a0, a1, a2
@end group
    .begin transform
M:  add a0, a1, a2
    .end transform
@group
N:  add a0, a1, a2
    .end no-transform
@end group
@end smallexample

The @code{ADD} opcodes at @code{L} and @code{N} in the outer
@code{no-transform} region both result in @code{ADD} machine instructions,
but the assembler selects an @code{ADD.N} instruction for the
@code{ADD} at @code{M} in the inner @code{transform} region.

The advantage of this style is that it works well inside macros which can
preserve the context of their callers.

The following directives are available:
@menu
* Schedule Directive::         Enable instruction scheduling.
* Longcalls Directive::        Use Indirect Calls for Greater Range.
* Transform Directive::        Disable All Assembler Transformations.
* Literal Directive::          Intermix Literals with Instructions.
* Literal Position Directive:: Specify Inline Literal Pool Locations.
* Literal Prefix Directive::   Specify Literal Section Name Prefix.
* Absolute Literals Directive:: Control PC-Relative vs. Absolute Literals.
@end menu

@node Schedule Directive
@subsection schedule
@cindex @code{schedule} directive
@cindex @code{no-schedule} directive

The @code{schedule} directive is recognized only for compatibility with
Tensilica's assembler.

@smallexample
@group
    .begin [no-]schedule
    .end [no-]schedule
@end group
@end smallexample

This directive is ignored and has no effect on @command{@value{AS}}.

@node Longcalls Directive
@subsection longcalls
@cindex @code{longcalls} directive
@cindex @code{no-longcalls} directive

The @code{longcalls} directive enables or disables function call
relaxation.  @xref{Xtensa Call Relaxation, ,Function Call Relaxation}.

@smallexample
@group
    .begin [no-]longcalls
    .end [no-]longcalls
@end group
@end smallexample

Call relaxation is disabled by default unless the @samp{--longcalls}
command-line option is specified.  The @code{longcalls} directive
overrides the default determined by the command-line options.

@node Transform Directive
@subsection transform
@cindex @code{transform} directive
@cindex @code{no-transform} directive

This directive enables or disables all assembler transformation,
including relaxation (@pxref{Xtensa Relaxation, ,Xtensa Relaxation}) and
optimization (@pxref{Xtensa Optimizations, ,Xtensa Optimizations}).

@smallexample
@group
    .begin [no-]transform
    .end [no-]transform
@end group
@end smallexample

Transformations are enabled by default unless the @samp{--no-transform}
option is used.  The @code{transform} directive overrides the default
determined by the command-line options.  An underscore opcode prefix,
disabling transformation of that opcode, always takes precedence over
both directives and command-line flags.

@node Literal Directive
@subsection literal
@cindex @code{literal} directive

The @code{.literal} directive is used to define literal pool data, i.e.,
read-only 32-bit data accessed via @code{L32R} instructions.

@smallexample
    .literal @var{label}, @var{value}[, @var{value}@dots{}]
@end smallexample

This directive is similar to the standard @code{.word} directive, except
that the actual location of the literal data is determined by the
assembler and linker, not by the position of the @code{.literal}
directive.  Using this directive gives the assembler freedom to locate
the literal data in the most appropriate place and possibly to combine
identical literals.  For example, the code:

@smallexample
@group
    entry sp, 40
    .literal .L1, sym
    l32r    a4, .L1
@end group
@end smallexample

can be used to load a pointer to the symbol @code{sym} into register
@code{a4}.  The value of @code{sym} will not be placed between the
@code{ENTRY} and @code{L32R} instructions; instead, the assembler puts
the data in a literal pool.

Literal pools are placed by default in separate literal sections;
however, when using the @samp{--text-@-section-@-literals}
option (@pxref{Xtensa Options, ,Command-line Options}), the literal
pools for PC-relative mode @code{L32R} instructions
are placed in the current section.@footnote{Literals for the
@code{.init} and @code{.fini} sections are always placed in separate
sections, even when @samp{--text-@-section-@-literals} is enabled.}
These text section literal
pools are created automatically before @code{ENTRY} instructions and
manually after @samp{.literal_position} directives (@pxref{Literal
Position Directive, ,literal_position}).  If there are no preceding
@code{ENTRY} instructions, explicit @code{.literal_position} directives
must be used to place the text section literal pools; otherwise,
@command{@value{AS}} will report an error.

When literals are placed in separate sections, the literal section names
are derived from the names of the sections where the literals are
defined.  The base literal section names are @code{.literal} for
PC-relative mode @code{L32R} instructions and @code{.lit4} for absolute
mode @code{L32R} instructions (@pxref{Absolute Literals Directive,
,absolute-literals}).  These base names are used for literals defined in
the default @code{.text} section.  For literals defined in other
sections or within the scope of a @code{literal_prefix} directive
(@pxref{Literal Prefix Directive, ,literal_prefix}), the following rules
determine the literal section name:

@enumerate
@item
If the current section is a member of a section group, the literal
section name includes the group name as a suffix to the base
@code{.literal} or @code{.lit4} name, with a period to separate the base
name and group name.  The literal section is also made a member of the
group.

@item
If the current section name (or @code{literal_prefix} value) begins with
``@code{.gnu.linkonce.@var{kind}.}'', the literal section name is formed
by replacing ``@code{.@var{kind}}'' with the base @code{.literal} or
@code{.lit4} name.  For example, for literals defined in a section named
@code{.gnu.linkonce.t.func}, the literal section will be
@code{.gnu.linkonce.literal.func} or @code{.gnu.linkonce.lit4.func}.

@item
If the current section name (or @code{literal_prefix} value) ends with
@code{.text}, the literal section name is formed by replacing that
suffix with the base @code{.literal} or @code{.lit4} name.  For example,
for literals defined in a section named @code{.iram0.text}, the literal
section will be @code{.iram0.literal} or @code{.iram0.lit4}.

@item
If none of the preceding conditions apply, the literal section name is
formed by adding the base @code{.literal} or @code{.lit4} name as a
suffix to the current section name (or @code{literal_prefix} value).
@end enumerate

@node Literal Position Directive
@subsection literal_position
@cindex @code{literal_position} directive

When using @samp{--text-@-section-@-literals} to place literals inline
in the section being assembled, the @code{.literal_position} directive
can be used to mark a potential location for a literal pool.

@smallexample
    .literal_position
@end smallexample

The @code{.literal_position} directive is ignored when the
@samp{--text-@-section-@-literals} option is not used or when
@code{L32R} instructions use the absolute addressing mode.

The assembler will automatically place text section literal pools
before @code{ENTRY} instructions, so the @code{.literal_position}
directive is only needed to specify some other location for a literal
pool.  You may need to add an explicit jump instruction to skip over an
inline literal pool.

For example, an interrupt vector does not begin with an @code{ENTRY}
instruction so the assembler will be unable to automatically find a good
place to put a literal pool.  Moreover, the code for the interrupt
vector must be at a specific starting address, so the literal pool
cannot come before the start of the code.  The literal pool for the
vector must be explicitly positioned in the middle of the vector (before
any uses of the literals, due to the negative offsets used by
PC-relative @code{L32R} instructions).  The @code{.literal_position}
directive can be used to do this.  In the following code, the literal
for @samp{M} will automatically be aligned correctly and is placed after
the unconditional jump.

@smallexample
@group
    .global M
code_start:
@end group
    j continue
    .literal_position
    .align 4
@group
continue:
    movi    a4, M
@end group
@end smallexample

@node Literal Prefix Directive
@subsection literal_prefix
@cindex @code{literal_prefix} directive

The @code{literal_prefix} directive allows you to override the default
literal section names, which are derived from the names of the sections
where the literals are defined.

@smallexample
@group
    .begin literal_prefix [@var{name}]
    .end literal_prefix
@end group
@end smallexample

For literals defined within the delimited region, the literal section
names are derived from the @var{name} argument instead of the name of
the current section.  The rules used to derive the literal section names
do not change.  @xref{Literal Directive, ,literal}.  If the @var{name}
argument is omitted, the literal sections revert to the defaults.  This
directive has no effect when using the
@samp{--text-@-section-@-literals} option (@pxref{Xtensa Options,
,Command-line Options}).

@node Absolute Literals Directive
@subsection absolute-literals
@cindex @code{absolute-literals} directive
@cindex @code{no-absolute-literals} directive

The @code{absolute-@-literals} and @code{no-@-absolute-@-literals}
directives control the absolute vs.@: PC-relative mode for @code{L32R}
instructions.  These are relevant only for Xtensa configurations that
include the absolute addressing option for @code{L32R} instructions.

@smallexample
@group
    .begin [no-]absolute-literals
    .end [no-]absolute-literals
@end group
@end smallexample

These directives do not change the @code{L32R} mode---they only cause
the assembler to emit the appropriate kind of relocation for @code{L32R}
instructions and to place the literal values in the appropriate section.
To change the @code{L32R} mode, the program must write the
@code{LITBASE} special register.  It is the programmer's responsibility
to keep track of the mode and indicate to the assembler which mode is
used in each region of code.

If the Xtensa configuration includes the absolute @code{L32R} addressing
option, the default is to assume absolute @code{L32R} addressing unless
the @samp{--no-@-absolute-@-literals} command-line option is specified.
Otherwise, the default is to assume PC-relative @code{L32R} addressing.
The @code{absolute-@-literals} directive can then be used to override
the default determined by the command-line options.

@c Local Variables:
@c fill-column: 72
@c End: