| Version 47 (modified by trac, 4 weeks ago) (diff) |
|---|
CLRadeonExtender Assembler Pseudo-Operations
The CLRX assembler accepts the almost pseudo-operations from GNU assembler. This chapter lists and explains standard pseudo-operations.
The CLRX assembler stores values greater than byte in the little-endian ordering.
List of the pseudo-operations
.32bit
This pseudo-operation should to be at begin of source. Choose 32-bit binaries (it has meaningful for the AMD Catalyst and GalliumCompute binary format). For AMD Catalyst OpenCL 1.2 and 2.0 format, it determines bitness of address. For GalliumCompute, it determines bitness of inner ELF binary. For ROCm binary format, bitness is ignored.
.64bit
This pseudo-operation should to be at begin of source. Choose 64-bit binaries (it has meaningful for the AMD Catalyst and GalliumCompute binary format). For AMD Catalyst OpenCL 1.2 and 2.0 format, it determines bitness of address. For GalliumCompute, it determines bitness of inner ELF binary. For ROCm binary format, bitness is ignored.
.abort
Aborts compilation.
.align, .balign
Syntax: .align ALIGNMENT[, [VALUE] [, LIMIT]]
Syntax: .balign ALIGNMENT[, [VALUE] [, LIMIT]]
Align current position to value of the first expression.
Value of that expression must be a power of two.
By default if section is writeable, fill 0's.
Second expression determines byte value that will be stored inside filling.
If third expression is not given, assembler accepts any length of the skip.
Third expression limits skip to own value. If any alignment needs to skip number
of the bytes greater than that value, then alignment will not be done.
If aligment will be done in .text section and second expresion will not be given, then
assembler fills no-operation instructions in that hole.
.altmacro
Enable alternate macro syntax. This mode enables following features:
- new macro substitution without backslash:
.macro test1 a b
.int a, b
.endm
test1 12,34 # put 12 and 34 integer value
- evaluating expression as string in macro arguments:
.macro stringize a,b,c
.string "a, b, c"
.endm
stringize %12|33, %43*5, %12-65 # generate string "45, 215, -53"
- new string quoting in macro arguments (by triagular brackets '<' and '>' or by single quote ' or double quote "). Also, enables new string escaping by '!'.
test1 <this is test !<!>> # put "this is test <>" string to first macro argument
test1 "this is test <>" # put "this is test <>" string to first macro argument
test1 'this is test !'' # put "this is test '" string to first macro argument
- local symbol names. 'local name' defines new unique symbol name for name. If any name will be occurred then that unique name will substituted.
local myName # myName substitutes new unique name
myName: # define new label with unique name
An alternate macro syntax does not disable any standard macro syntax features likes macro substitution via backslashes, '\@'.
.amd
This pseudo-operation should to be at begin of source. Choose AMD Catalyst OpenCL 1.2 program binary format.
.amdcl2
This pseudo-operation should to be at begin of source. Choose AMD Catalyst new (introduced for OpenCL 2.0 support) program binary format.
.amd3
This pseudo-operation should to be at begin of source. Choose ROCm LLV10 binary format with metadata V3 format - new AMD OpenCL format for Navi (GCN1.5).
.arch
Syntax: .arch ARCHITECTURE
This pseudo-operation should to be at begin of source. Set GPU architecture. One of following architecture can be set: SI, VI, CI, VEGA, VEGA20, Navi, NaviDL, GFX6, GFX7, GFX8, GFX9, GFX906, GFX10, GFX1011, GCN1.0, GCN1.1, GCN1.2, GCN1.4, GCN1.4.1, GCN1.5, GCN1.5.1.
.ascii
Syntax: .ascii "STRING",....
Emit ASCII string. This pseudo-operations does not add the null-terminated character. If more than one string will be given then all given string will be concatenated.
.asciz
Syntax: .asciz "STRING",....
Emit ASCII string. This pseudo-operations adds the null-terminated character. If more than one string will be given then all given string will be concatenated.
.balignw, .balignl
Syntax: .balignw ALIGNMENT[, [VALUE] [, LIMIT]]
Syntax: .balignl ALIGNMENT[, [VALUE] [, LIMIT]]
Refer to .align. .balignw treats fill value as 2-byte word. .balignl treats
fill value as 4-byte word.
.buggyfplit
Enable old and buggy behavior for floating point literals and constants.
.byte
Syntax: .byte ABS-EXPR,....
Emit byte values. If any expression is empty then the assembler stores 0 and warns about empty expression. If expression will give a value that can not be stored in byte then the assembler warn about that.
.data
Go to .data section. If this section doesn't exist assembler create it.
.double
Syntax: .double DOUBLE-VAL,...
Put double-precision floating point values into current section. If no value between comma then the assembler stores 0 and warn about no value. This pseudo-operation accepts only double precision floating point literals.
.else
Part of the if-endif clause. Refer to .if pseudo-operation.
Code between .else and .endif will be performed if all previous conditions was not
satisified. Otherwise code will be skipped.
.elseifXXX
Syntax: .elseif ABS-EXPR
Syntax: .elseif32
Syntax: .elseif64
Syntax: .elseifarch ARCHITECTURE
Syntax: .elseifb STRING
Syntax: .elseifc STRING1, STRING2
Syntax: .elseifdef SYMBOL
Syntax: .elseifeq ABS-EXPR
Syntax: .elseifeqs "STRING1","STRING2"
Syntax: .elseiffmt BINFMT
Syntax: .elseifge ABS-EXPR
Syntax: .elseifgpu GPUDEVICE
Syntax: .elseifgt ABS-EXPR
Syntax: .elseifle ABS-EXPR
Syntax: .elseiflt ABS-EXPR
Syntax: .elseifnarch ARCHITECTURE
Syntax: .elseifnb STRING
Syntax: .elseifnc STRING1, STRING2
Syntax: .elseifndef SYMBOL
Syntax: .elseifne ABS-EXPR
Syntax: .elseifnes "STRING1","STRING2"
Syntax: .elseifnfmt BINFMT
Syntax: .elseifngpu GPUDEVICE
Syntax: .elseifnotdef SYMBOL
Part of the if-endif clause. Refer to .if pseudo-operation.
Code between .else and .endif will be performed if all previous conditions was not
satisified and condition to this pseudo-operation was satisfied. Otherwise code will be
skipped.
.end
Ends source code compilation at this point.
.endif
Finish if-endif clause.
.endm, .endmacro
Finish macro definition
.endr, .endrept
Finish code of repetition.
.ends, .endscope
Close visibility's scope.
.enum
Syntax: .enum [>[STARTPOS],] SYMBOL,....
Simplify defining the enumerations. For every symbol,
define symbol with value of enumeration counter and increase an enumeration counter.
Defined symbols can not be assigned later. Optional > with optional STARTPOS sets
enumeration counter to STARTPOS value or zero (if no STARTPOS given).
Every scope has own enumeration counter. This features simplify
joining enumerations with scopes.
Examples:
.enum OK,BADFD,FATAL # define OK=0, BADFD=1, FATAL=2
.enum >8, BitOne,BitTwo # define BitOne=8, BitTwo=9
.enum HALO # define HALO=10
.scope Result
# enum counter is zero in this scope
.enum NONE,FULL,INVALID # NONE=0, FULL=1, INVALID=2
.ends
.enum >, myzero # myzero=0
.equ, .set
Syntax: .equ SYMBOL, EXPR|REG
Syntax: .set SYMBOL, EXPR|REG
Define symbol with specified value of the expression or register or register range given in second operand. Symbol defined by using these pseudo-operations can be redefined many times.
.equiv
Syntax: .equiv SYMBOL, EXPR|REG
Define symbol with specified value of the expression or register or register range
given in second operand given in second operand.
Symbol defined by using .equiv can not be redefined. If symbol was already defined
this pseudo-operations causes an error.
.eqv
Syntax: .eqv SYMBOL, EXPR|REG
Define symbol with specified expression or register or register range
given in second operand given in second operand.
Symbol defined by using .eqv can not be redefined.
If symbol was already defined this pseudo-operations causes an error.
The expression of symbol will be evaluated any time when symbol will be used.
This feature allow to change value of symbol indirectly by changing value of the symbols
used in expression. Example:
a = 3
b = 5
.eqv currentValue, a+b # now currentValue is equal to 8
.int currentValue
b = 7
.int currentValue # we changed b to 7, so we put 10
.err
This pseudo-operation causes error at point where is encountered.
.error
Syntax: .error "STRING"
This pseudo-operation causes error and print error given in string.
.exitm
This pseudo-operation can be used only inside macro definitions. It does early exiting from macro. Useful while using conditions inside a macros.
.extern
Syntax: .extern SYMBOL,...
Set symbol as external. Ignored by GNU as and CLRX assembler.
.fail
Syntax: .fail ABS-EXPR
If value of the expression is greater or equal to 500 then assembler print warnings. Otherwise assembler print error and fails.
.file
This pseudo-operation is ignored by CLRX assembler.
.fill, .fillq
Syntax: .fill REPEAT[, [SIZE] [, VALUE]]
Syntax: .fillq REPEAT[, [SIZE] [, VALUE]]
Emit value many times. First expression defines how many times value will be stored.
Second expression defines how long is value. Third expression defines value to be stored.
If second expression is not given, then assembler assumes that is byte value.
If third expression is not given then assembler stores 0's. Assembler takes only
lowest 4-bytes of the value, and if size of value is greater than 4 then more significant
bytes will be assumed to be 0. The .fillq takes all bytes of the 64-bit value.
If value doesn't fit to size of value then it will be truncated.
Size of value can be 0. First expression too.
.float, .single
Syntax: .float FLOAT-VAL,...
Syntax: .single FLOAT-VAL,...
Emit single-precision floating point values. If no value between comma then an assembler stores 0 and warn about no value. This pseudo-operation accepts only single precision floating point literals.
.for
Syntax: .for SYMBOL=INITVALUE, COND-EXPR, NEXT-EXPR
Open 'for' repetition. Before repetition, SYMBOL is initialized by INITVALUE.
For every repetition SYMBOL value will be replaced by value evaluted by NEXT-EXPR.
The code between this pseudo-operation and .endr will be repeated
until COND-EXPR returns zero. Example:
.for x=1,x<16,x+x
.int x
.endr
generates:
.int x # x=1
.int x # x=2
.int x # x=4
.int x # x=8
.format
Syntax: .format BINFORMAT
This pseudo-operation should to be at begin of source. Choose binary format. Binary can be one of following list:
amd,catalyst- AMD Catalyst OpenCL 1.2 binary formatamdcl2- AMD Catalyst new (introduced for OpenCL 2.0) binary formatgallium- the GalliumCompute binary formatrocm- the ROCm binary formatraw- rawcode (raw program instructions)
.gallium
This pseudo-operation should to be at begin of source. Choose GalliumCompute OpenCL program binary format.
.get_64bit
Syntax: .get_64bit SYMBOL
Store 64-bitness in specified symbol. Store 1 to symbol if 64-bit mode enabled.
.get_arch
Syntax: .get_arch SYMBOL
Store GPU architecture identifier to symbol. List of architecture ids:
| Id | Description |
|---|---|
| 0 | GCN1.0 (Pitcairn, Tahiti) |
| 1 | GCN1.1 (Bonaire, Hawaii) |
| 2 | GCN1.2 (Tonga, Fiji, Ellesmere) |
| 3 | VEGA (AMD RX VEGA) |
| 4 | VEGA20 (GFX906) |
| 5 | NAVI (RX 5700) |
| 6 | NAVI DL |
.get_format
Syntax: .get_format SYMBOL
Store binary format identifier to symbol. List of format ids:
| Id | Description |
|---|---|
| 0 | AMD OpenCL 1.2 binary format |
| 1 | Gallium Compute binary format |
| 2 | Raw code |
| 3 | AMD OpenCL 2.0 (new driver) binary format |
| 4 | ROCm binary format |
.get_gpu
Syntax: .get_gpu SYMBOL
Store GPU device identifier to symbol. List of GPU device ids:
| Id | Description |
|---|---|
| 0 | Cape Verde (Radeon HD 7700) |
| 1 | Pitcairn (Radeon HD 7850) |
| 2 | Tahiti (Radeon HD 7900) |
| 3 | Oland |
| 4 | Bonaire (Radeon R7 260) |
| 5 | Spectre |
| 6 | Spooky |
| 7 | Kalindi |
| 8 | Hainan |
| 9 | Hawaii (Radeon R9 290) |
| 10 | Iceland |
| 11 | Tonga (Radeon R9 285) |
| 12 | Mullins |
| 13 | Fiji (Radeon Fury) |
| 14 | Carrizo |
| 15 | Dummy |
| 16 | Goose |
| 17 | Horse |
| 18 | Stoney |
| 19 | Ellesmere (Radeon RX 470/480) |
| 20 | Baffin (Radeon 460) |
| 21 | gfx804 |
| 22 | gfx900 (Radeon RX VEGA) |
| 23 | gfx901 (Radeon RX VEGA) |
| 24 | gfx902 (Radeon RX VEGA) |
| 25 | gfx903 (Radeon RX VEGA) |
| 26 | gfx904 (Radeon RX VEGA) |
| 27 | gfx905 (Radeon RX VEGA) |
| 28 | gfx906 (Radeon VEGA 20) |
| 29 | gfx907 (Radeon VEGA 20 ???) |
| 30 | gfx1000 |
| 31 | gfx1010 |
| 32 | gfx1011 |
.get_policy
Syntax: .get_version SYMBOL
Store current CLRX policy version to symbol. Version stored as integer in form:
major_version*10000 + minor_version*100 + micro_version.
.get_version
Syntax: .get_version SYMBOL
Store CLRadeonExtender version to symbol. Version stored as integer in form:
major_version*10000 + minor_version*100 + micro_version.
.global, .globl
Syntax: .global SYMBOL,...
Syntax: .globl SYMBOL,...
Indicates that symbols will be a global. Global symbol can be used across many object files. The CLRX assembler can expose global symbols at output binary if adding sybols will be enabled.
.gpu GPUDEVICE
Syntax: .gpu GPUDEVICE
This pseudo-operation should to be at begin of source. Set GPU device. One of following device type can be set: Baffin, Bonaire, CapeVerde, Carrizo, Dummy, Ellesmere, Fiji, GFX700, GFX701, GFX801, GFX802, GFX803, GFX804, GFX900, GFX901, GFX902, GFX903, GFX904, GFX905, GFX906, GFX907, Goose, Hainan, Hawaii, Horse, Iceland, Kalindi, Mullins, Oland, Pitcairn, Polaris10, Polaris11, Polaris12, Polaris20, Polaris21, Polaris22, Raven, Spectre, Spooky, Stoney, Tahiti, Tonga, Topaz, Vega10, Vega11, Vega12, Vega20, GFX1000, GFX1010, GFX1011, Navi10.
.half
Syntax: .half HALF-VAL,...
Emit half-precision floating point values. If no value between comma then the assembler stores 0 and warn about no value. This pseudo-operation accepts only half precision floating point literals.
.hword, .short
Syntax: .hword ABS-EXPR,.... Syntax: .short ABS-EXPR,....
Emit 2-byte word values. If any expression is empty then the assembler stores 0 and warns about empty expression. If expression will give a value that can not be stored in 2-byte word then the assembler warn about that.
.ifXXX
Syntax: .if ABS-EXPR
Syntax: .if32
Syntax: .if64
Syntax: .ifarch ARCHITECTURE
Syntax: .ifb STRING
Syntax: .ifc STRING1, STRING2
Syntax: .ifdef SYMBOL
Syntax: .ifeq ABS-EXPR
Syntax: .ifeqs "STRING1","STRING2"
Syntax: .iffmt BINFMT
Syntax: .ifge ABS-EXPR
Syntax: .ifgpu GPUDEVICE
Syntax: .ifgt ABS-EXPR
Syntax: .ifle ABS-EXPR
Syntax: .iflt ABS-EXPR
Syntax: .ifnarch ARCHITECTURE
Syntax: .ifnb STRING
Syntax: .ifnc STRING1, STRING2
Syntax: .ifndef SYMBOL
Syntax: .ifne ABS-EXPR
Syntax: .ifnes "STRING1","STRING2"
Syntax: .ifnfmt BINFMT
Syntax: .ifngpu GPUDEVICE
Syntax: .ifnotdef SYMBOL
Open if-endif clause. After that pseudo-op can encounter .elseifXX or .else
pseudo-operations. If condition of that pseudo-operations is satisfied then
code will be performed.
List of the .if kinds:
.if- perform code if value is not zero..if32- perform code if 32-bit binary..if64- perform code if 64-bit binary..ifarch- perform code if specified architecture was set..ifb- perform code if all character after name of this pseudo-op is blank..ifc- compare two unquoted strings. If are equal then perform code..ifdef- perform code if symbol was defined.ifeq- perform code if value is zero..ifeqs- perform code if two quoted string are equal..iffmt- perform code if specified binary format was set..ifge- perform code if value is greater or equal to zero..ifgpu- perform code if specified GPU device type was set..ifgt- perform code if value is greater than zero..ifle- perform code if value is less or equal to zero..iflt- perform code if value is less than zero..ifnarch- perform code if specified architecture was not set..ifnb- perform code if any character after name of this pseudo-op is not blank..ifnc- compare two unquoted strings. If are not equal then perform code..ifndef- perform code if symbol was not defined..ifne- perform code if value is not zero..ifnes- perform code if two quoted string are not equal..ifnfmt- perform code if specified binary format was not set..ifngpu- perform code if specified GPU device type was not set..ifnotdef- perform code if symbol was not defined.
NOTE: For ifarch or .ifnarch:
The assembler assumes that VEGA20 is this same ISA as VEGA10 (GFX9)
and if you set VEGA20 GPU architecture and if you put GFX9 architecture in .ifarch
then code for this clause will be assembled (or will not be assembled for .ifnarch).
.incbin
Syntax: .incbin FILENAME[, [OFFSET] [, COUNT]]
Append the binary file into currenct section. If file not found in the current directory then assembler searches file in the include paths. If file not found again then assembler prints error. Second optional argument defines offset (how many bytes should to be skipped). By default assembler begins appending from first byte. Third argument defines maximum number bytes to append. By default all data from binary will be appended.
.include
Syntax: .include "FILENAME"
Include new source code file and begins assemblying from this file. The assembler automatically returns to previous file if encounters end of the that file. If file not found in the current directory then assembler searches file in the include paths. If file not found again then assembler prints error.
.irp
Syntax: .irp NAME, STRING,...
Syntax: .irp NAME STRING ...
Begin repetition with the named variable. Nth occurrence of name will be replaced by
nth string. The name of variable must be preceded by \ character.
Sample usage:
.irp reg v0,v1,v4,v5
v_mul_f32 v10,\reg
.endr
produces:
v_mul_f32 v10,v0
v_mul_f32 v10,v1
v_mul_f32 v10,v4
v_mul_f32 v10,v5
Alternate syntax does not require a comma separator between strings and variable name.
Rules regarding to substituting variables are same as in macro substitution. Refer to
.macro pseudo-operation.
.irpc
Syntax: .irpc NAME, STRING
Begin repetition with the named variable.Nth occurrence of name will be replaced by
nth character of the string (second operand).
The name of variable must be preceded by \ character.
.irp reg 0145
v_mul_f32 v10,v\reg
.endr
produces:
v_mul_f32 v10,v0
v_mul_f32 v10,v1
v_mul_f32 v10,v4
v_mul_f32 v10,v5
Rules regarding to substituting variables are same as in macro substitution. Refer to
.macro pseudo-operation.
.int, .long
Syntax: .int ABS-EXPR,....
Syntax: .long ABS-EXPR,....
Emit 4-byte word values. If any expression is empty then the assembler stores 0 and warns about empty expression. If expression will give a value that can not be stored in 4-byte word then the assembler warn about that.
.kernel
Syntax: .kernel KERNELNAME
Create new kernel with default section or move to an existing kernel and its default section. Types of section which can to be belonging to kernel depends on the binary format.
.lflags
This pseudo-operation is ignored by CLRX assembler.
.ln, .line
These pseudo-operations are ignored by CLRX assembler.
.local
Syntax: .local SYMBOL,...
Indicates that symbols will be a local. Local symbol are not visible in other binary objects and they can be used only locally.
.macro
Syntax: .macro MACRONAME, ARG,...
Syntax: .macro MACRONAME ARG ...
Begins macro definition. The macro is recorded code that can be used later in source code.
Macro can accepts one or more arguments that will be substituted in its body.
Occurrence of the argument must be preceded by \ character.
Special case of substitution is \@ that replaced by number of the macro substitution.
Sometimes substitution must be joined with other name. In this case '()` can be used to
split argument occurrence and other name. Example:
.macro macro1, x
.set \x\()Symbol, 10 # set symbol Xsymbol to value 10
.endm
macro1 Linux # define LinuxSymbol=10
That substitution is useful to create labels and symbols that can not be redefined. Value of the argument is string. Optionally, argument can has the default value which will be used if no argument value is not given in a macro call.
List of the argument definition:
arg1- simple argumentarg1=value- argument with default valuearg1:req- required argumentarg:vararg- variadic argument, get rest of the arguments and treats them as single (must be last)arg1:vararg:req- required argument and variadic argument
A macro calling is easy:
somemacro
somemacro1 1,2,4
somemacro1 1 2 3
/* use variadic argument */
.macro putHeaderAndData x,data:vararg
.byte \x&0x3f, \data
.endm
putHeaderAndData 0xfe,1,4,6,7,110
Some simple examples:
.macro putTwoBytes value1,value2
.byte \value1+\value2, \value2-\value1
.endm
putTwoBytes 4,7 # generates .byte 11,3
.macro putTwoBytes2 value1=100,value2
.byte \value1+\value2, \value2-\value1
.endm
putTwoBytes2 ,7 # generates .byte 107,93
The argument substitution works inside double-quoted strings:
.macro putVersionInfo major,minor,micro
.ascii "CLRadeonExtender \major\().\minor\().\micro"
.endm
putVersionInfo 1,2,3 # generates string "CLRadeonExtender 1.2.3"
It is possible to create the nested definitions like that:
.macro generators m,value1,value2
.macro gen\m x1,x2 # define genM with two arguments
/* use first and second operand of parent macro, and use two
* operands from this macro */
.int (\x1+\x2)*(\value1+\value2)
.endm
.endm
/* define genF1 */
generators F1,2,3
/* use genF1, generates (10+12)*(2+3) -> 22*5 -> 110 */
genF1 10,12
If macro argument starts with '\%' then rest of macro argument will be evaluated as expression and its results will be stored in string form (likewise '%EXPR' evaluation if alternate macro syntax is enabled).
.macrocase
Enable ignoring letter's case in macro names (default behaviour).
.main
Go to main binary over binary of the kernel.
.noaltmacro
Disables alternate macro syntax.
.nobuggyfplit
Disable old and buggy behavior for floating point literals and constants.
.nomacrocase
Disable ignoring letter's case in macro names.
.nooldmodparam
Disable old modifier parametrization that accepts only 0 and 1 values (to 0.1.5 version).
.nowave32
Disable wavefront size as 32 elements (apply only for GFX10 devices). Use default 64 element wavesize.
.octa
Syntax: .octa OCTA-LITERAL,...
Emit 128-bit word values. If no value between comma then the assembler stores 0 and warn about no value. This pseudo-operation accepts only 128-bit word literals.
.offset, .struct
Syntax: .offset ABS-EXPR
Syntax: .struct ABS-EXPR
Set the output counter to some place in absolute section. Useful to defining fields of the structures.
.oldmodparam
Enable old modifier parametrization that accepts only 0 and 1 values (to 0.1.5 version) for compatibility.
.org
Syntax: .org EXPRESSION
Set the output counter to place defined by expression. Expression can point to some place in the code or some place to absolute section (absolute value). If a expression points to some place of a code, moving backwards is illegal.
.p2align
Syntax: .p2align POWOF2ALIGN[, [VALUE] [, LIMIT]]
Refer to .align. First argument is power of two of the alignment instead of
same alignment.
.policy
Syntax: .policy VERSION
Set current CLRX assembler's policy version. Version stored as integer in form:
major_version*10000 + minor_version*100 + micro_version. This number controls
behaviour of the same assembler for some things like SGPRs counting. Refer to
Asssembler policy.
Syntax: .print "STRING"
Print string into standard output (or defined print output).
.purgem
Syntax: .purgem MACRONAME
Undefine macro MACRONAME.
.quad
Syntax: .quad ABS-EXPR,....
Emit 8-byte word values. If any expression is empty then the assembler stores 0 and warns about empty expression.
.rawcode
This pseudo-operation should to be at begin of source. Choose raw code (same processor's instructions).
.regvar
Syntax: .regvar REGVAR:REGTYPE:REGSNUM, ...
Define new register variable (UNIMPLEMENTED).
.rept
Syntax: .rept ABS-EXPR
Open repetition. The code between this pseudo-operation and .endr will be repeated
number given in first argument. Zero value is legal in first argument. Example:
.rept 3
v_nop
.endr
generates:
v_nop
v_nop
v_nop
.rocm
This pseudo-operation should to be at begin of source. Choose ROCm program binary format.
.rodata
Go to .rodata section. If this section doesn't exist assembler create it.
.sbttl, .tittle, .version
Syntax: .sbttl "STRING"
Syntax: .title "STRING"
Syntax: .version "STRING"
These pseudo-operations are ignored by CLRX assembler.
.scope
Syntax .scope [SCOPENAME]
Open visbility's scope (if no name specified, then temporary scope). The labels (except local labels), symbols, scopes and regvars are defined inside scopes and visible inside them. The assembler create always global scope at begin. If scope doesn't exists then will be created. The nested scopes are allowed even if parent scope is temporary scope. The opened scope has parent that is previous scope. Temporary scopes exists until first close.
.section
Syntax: .section SECTIONNAME[, "FLAGS"[, @TYPE]] [align=ALIGN]
Go to specified section SECTIONNAME. If section doesn't exist assembler create it. Second optional argument set flags of the section. Can be from list:
a- allocatable sectionw- writeable section in filex- executable section
Third optional argument set type of the section. Default section's type depends on the binary format. Type can be one of following type:
progbits- program datanote- informations about program or other thingsnobits- doesn't contain data (only occupies space)
The last attribute called 'align' set up section aligmnent.
.size
Syntax: .size SYMBOL, ABS-EXPR
Set size of symbol SYMBOL. Currently, this feature of symbol is not used by the CLRX assembler.
.skip, .space
Syntax: .skip SIZE-EXPR[, VALUE-EXPR] Syntax: .space SIZE-EXPR[, VALUE-EXPR]
Likewise as .fill, this pseudo-operation emits value (but byte) many times.
First expression determines how many values should to be emitted, second expression
determines what byte value should to be stored. If second expression is not given
then assembler stores 0's.
.string, .string16, .string32, .string64
Syntax: .string "STRING",....
Syntax: .string16 "STRING",....
Syntax: .string32 "STRING",....
Syntax: .string64 "STRING",....
Emit ASCII string. This pseudo-operations adds the
null-terminated character. If more than one string will be given then all given
string will be concatenated.
.string16 emits string with 2-byte characters.
.string32 emits string with 4-byte characters.
.string64 emits string with 8-byte characters.
Characters longer than 1 byte will be zero expanded.
.text
Go to .text section. If this section doesn't exist assembler create it.
.undef
Syntax: .undef SYMBOL
Undefine symbol. If symbol already doesn't exist then assembler warns about that.
.unusing
Syntax: .unusing [SCOPEPATH]
Stop using all objects (labels, symbols, regvar) in the specified scope or all previously used scopes (by '.using').
.using
Syntax: .using SCOPEPATH
Start use all objects (labels, symbols, scopes, regvar) in the specified scope. All objects (if they names is not hidden by object in current scope) will be visible directly including usings from specified scope. Searching object in current level of scope's stack is recursive and it traverse through 'usings' until object will be found. Assembler start searching object begins from the last declared 'using' to first 'using'.
.warning
Syntax: .warning "STRING"
Print warning message specified in first argument.
.wave32
Set wavefront size as 32 elements (apply only for GFX10 devices).
.weak
Syntax: .weak SYMBOL,...
Indicates that symbols will be weak. Currently, unused feature of the symbol by the CLRX assembler.
.while
Syntax: .while COND-EXPR
Open 'while' repetition.The code between this pseudo-operation and .endr will be repeated
until COND-EXPR returns zero. Example:
x=1
.while x<16
.int x
x=x+x
.endr
generates:
.int x # x=1
.int x # x=2
.int x # x=4
.int x # x=8
.word
Syntax: .word ABS-EXPR,....
Emit processor's word values. If any expression is empty then the assembler stores 0 and warns about empty expression. If expression will give a value that can not be stored in processor's word then the assembler warn about that.
Processor's word has always 4 bytes.
