-Wparentheses
In some cases, if you ignore the brackets
,
Warning
.
-WTemplate-Debugging
In
C
Used in the program
Template
when
,
If debugging
(Debugging)
Not completely effective
,
Warning
(
only use for
C ).
-Wall
Combined with all of the above
`-W '
Option
.
Usually we recommend to avoid these warned usage, we believe
,
The use of proper combination of macros can easily avoid these usage.
The remaining `-w ... 'option is not included in` -wall', because we believe that these compiler warned program structures can be reasonably used in the "clean" program.
-Wtraditional
If some program structure is in a tradition
C
Performance and
ANSI C
different
,
Warning
.
*
Macro ginseng appears in a string constant in the macro body
.
Tradition
C
Alternative macro
,
and
ANSI C
Then depend on part of a constant
.
*
A function in block
(Block)
Declaration is external
,
But call after the end of the block
.
*
Switch
The operability type of the statement is
Long.
-Wshadow
Once some local variables are blocked, another part variable
,
Warning
.
-WID-CLASH-LEN
Once the two determined identifiers have the same top
Len
Character
,
Warning
.
He can help you develop some obsolete
,
Procedure compiled on the compiler of the brain
.
-Wpointer-Arith
Any statement relies on the size of the function type
(size)
or
Void
Type size
,
Warning
GNU C
For convenience of calculation
Void *
Pointer and function pointer
,
That is defined as the size of these types
1.
-WCAST-QAL
Once a pointer is forced to transform, when the type modifier is removed
,
Warning
.
E.g
,
If
Const char *
Forced to convert to ordinary
Char *
Time
,
Warning will appear
.
-WCAST-ALIGN
Once a pointer type is forced to convert
,
Alignment to the address required for the target
(Alignment)
increase
,
Warning
.
E.g
,
Some machines can only be
2
or
4
Access integer on the byte boundary
,
If you put on this model
Char *
Forced conversion
Int *
Types of
,
Warning
.
-WWRITE-STRINGS
The type of string constant is specified is
Const char [length],
therefore
,
Copy this address to
Non-const char *
The pointer will produce a warning
.
These warnings can help you find out the code that is trying to write a critical quantity during compilation.
,
But you must use very carefully in the statement and intra shape
CONST,
Otherwise they can only bring trouble
;
So we didn't let
`-Wall '
Provide these warnings
.
-WConversion
If the type conversion caused by a function is different, the type conversion is different when there is no function.
,
Warning
.
Here, the mutual conversion of the fixed point and floating point number
,
Change the width or symbol of the fixed point number
,
Unless they and default statements
(Default Promotion)
the same
.
-Waggregate-return
If you define or call the return structure or a combined function
,
Warning
(
From language perspective you can return an array
,
However, it will cause warnings
.)
-Wstrict-prototypes
If the function's declaration or definition does not point out parameter types
,
Warning
(
If the forward reference description of the function points to the parameter type
,
Allows the use of old-style function definitions
,
Can't produce warning
.)
-Wmissing-prototypes
If there is no pre-declare function, the full-term function is defined.
,
Warning
.
Even if the function defines the original shape, it will generate this warning.
.
His purpose is to check the global function that is not declared in the header file.
.
-Wmissing-declarations
If there is no pre-declaration, it defines a full-class function.
,
Warning
.
Even if the function defines the original shape, it will generate this warning.
.
The purpose of this option is to check the global function that is not declared in the header file.
.
-Wredundant-decls
If you define multiple declarations in the same visible domain
,
Warning
,
Even if these repetitive declarations are valid and have no difference
.
-Wnested-externs
If a certain
Extern
Declaration appears inside the function
,
Warning
.
-Wenum-Clash
A warning for the conversion between different enumeration types
(
Only applies to
C ).
-Wlong-long
If used
Long Long
Types issued a warning
.
This warning is the default
.
use
`-Wno-long-long '
Options to prevent this warning
. `-Wlong-long '
with
`-Wno-long-long '
only at
`-pedant
Under the way
.
-WOverLoaded-Virtual
(
Only applies to
C .)
In inheritance
,
Definition of virtual functions must match the type feature declared in the virtual function in the base class
(TYPE SIGNATURE).
When the inheritance is declared a function
,
It may be a wrong attempt to define a virtual function
,
Use this option to generate warnings
:
that is
,
When a function and the virtual function in the base class are the same name
,
However, the type characteristics do not meet any virtual functions of the base class.
,
The compiler will issue a warning
.
-Winline
If a function cannot embed inner
(inline),
Whether it is a statement
inline
Or is it specified
-finline-functions
Option
,
The compiler will issue a warning
.
-Wrror
Vision warning is wrong
;
Any warning will give up compilation
.
Debugging option (DEBUGGING OPTION)
GNU CC
Have a lot of special options
,
Can debug users' programs
,
Can also
GCC
Tired
:
-g
Local format in operating system
(Stabs, Coff, XCoff,
or
DWARF).
Generate debugging information
GDB
Can use these debugging information
.
On most of the system using the Stabs format, `-g 'option launches additional debugging information that only GDB is used; this information makes the GDB debugging effect, but it is possible to cause other debugger crashes, or refuse to read the program. If You determine if you want to generate additional information, use `-gstabs ',` -gstabs', `-gxcoff ',` -gxcoff', `-gdwarf ', or` -gdwarf' (see below).
Unlike most C compilers, the GNU CC allows for use in the `-g 'and` -o' option. Optimized code occasionally manufactures some surprises: some declared variables do not exist at all; the control flow directly to It is expected that some statements are not executed because the calculation results are constant or have been determined; some statements are executed elsewhere because they have been moved outside.
However, it proves that the output of debug optimization is possible. It is reasonable to use an optimizer to use an optimizer that may contain errors.
If the GNU CC supports outputting multiple debugging information, the following options are very useful.
-ggdb
Local format
(
If supported
)
Output debugging information
,
As much as possible
GDB
Expand
.
-gstabs
In
Stabs
format
(
If supported
)
Output debugging information
,
Not included
GDB
Expand
.
This is most
BSD
System
DBX
Use format
.
-gstabs
In
Stabs
format
(
If supported
)
Output debugging information
,
Use only
Gnu
Debugger (GDB)
Understandable
Gnu
Expand
.
Using these extensions may result in other debuggers crashing or refusing to read
.
-gcoff
In
Coff
format
(
If supported
)
Output debugging information
.
This is at
System v
Most of the fourth edition
System v
System
SDB
Use format
.
-gxcoff
In
XCOFF
format
(
If supported
)
Output debugging information
.
this is
IBM RS / 6000
System
DBX
Format used by the debugger
.
-gxcoff
In
XCOFF
format
(
If supported
)
Output debugging information
,
Use only
Gnu
Debugger
(GDB)
Understandable
Gnu
Expand
.
Using these extensions may result in other debuggers crashing or refusing to read
.
-gdwarf
In
DWARF
format
(
If supported
)
Output debugging information
.
This is most
System v
Fourth edition system
SDB
Use format
.
-gdwarf
In
DWARF
format
(
If supported
)
Output debugging information
,
Use only
Gnu
Debugger
(GDB)
Understandable
Gnu
Expand
.
Using these extensions may result in other debuggers crashing or refusing to read
.
-glevel -ggdblevel -gstabslevel -gcofflevel -gxcofflevel
-gdwarflevel
Request to generate debugging information
,
At the same time
Level
Point out how much information
.
default
Level
Value
2.
Level 1 outputs the minimum amount of information, only enough in the program that does not plan to debug, include the description of the function and external variable, but there is no local variable and line number information.
Level 3 contains more information, such as all macro definitions that appear in the program. When you use the `-g3 'option, some debuggers support macro expansion.
-p
Generate additional code
,
Output
Profile
information
,
Analysis program
PROF
use
.
-pg
Generate additional code
,
Output
Profile
information
,
Analysis program
GPROF
use
.
-A
Generate additional code
,
Used to output basic blocks
(Basic Block)
of
Profile
information
,
It records the number of executions of each basic block
,
For example
TCOV
Program analysis of this class
.
But pay attention
,
This data format is not
TCOV
expected
.
finally
GNU Gprof
These data will be processed
.
-AX
Generate additional code
,
Use
'bb.in'
Document reading basic block
Profile
parameter
,
Put
Profile
The result is written
'bb.out'
file
. `bb.in '
Contains a list of functions
.
Once a function in the list is entered
, PROFILE
Operation begins
,
After leaving the outermost function
, PROFILE
End of operation
.
In
`- '
Exclude the function of the preceded name
Profile
Outside operation
.
If the function name is not unique
,
It can write
`/path/filename.d:functionname '
Come to clarify
. `bb.out '
Will list some valid file names
.
These four function names have special meaning
: `__BB_JUMPS__ '
Lead to jump
(JUMP)
Frequency writing
`bb.out '.` __bb_trace__'
Resulting in the basic block sequence through the pipe
`gzip ',
Output
`bbtrace.gz '
file
. `__BB_HIDECALL__ '
Leading from tracking
(Trace)
Exclude
Call
instruction
. `__BB_SHOWRET__ '
Resulting in tracking instructions
.
-dletters
When compiled,
in
letters
Specified timing
(DUMP).
Used to debug compiler
.
Most dump file names add words to the source file name
(
E.g
`foo.c.rtl '
or
`foo.c.jump ').
-dm
Dump all macro definitions at the end of the pretreatment
,
No output to the file
.
-DN
Dump all macro names at the end of the pretreatment
.
-DD
Dump all macro definitions at the end of the pretreatment
,
Eat on normal output
.
-dy
Parsing
(PARSE)
Time to transfer debugging information in standard error
.
-dr
RTL
Dump after the stage
`file.rtl '.
-DX
Generate only the function
RTL,
Not compiling
.
Usually and
`r '
Join
.
-dj
After the first jump optimization
`file.jump '.
-DS
CSE
Including sometimes following
CSE
Rear jump optimization
)
After dump
`file.cse '.
-DL
Recurns and then dump
`file.loop '.
-dt
the second time
CSE
deal with
(
Including sometimes following
CSE
Rear jump optimization
)
After dump
`file.cse2 '.
-DF
Process analysis
(Flow Analysis)
After dump
`file.flow '.
-DC
Instruction combination
(Instruction Combination)
After dump
`file.combine '.
-DS
First instruction arrangement
(instruction schedule)
After dump
`file.sched '.
-DL
Dump after the local register is allocated
`file.lreg '.
-dg
The global register is allocated after dump
`file.greg '.
-dr
Second command arrangement
(instruction schedule)
After dump
`file.sched2 '.
-dj
The last jump optimization is dump
`file.jump2 '.
-DD
Postpone branch schedule
Delayed Branch Scheduling
After dump
`file.dbr '.
-DK
register
-
The stack is transferred after transition to
`file.stack '.
-DA
Generate all dumps
.
-dm
After running
,
In standard error display memory usage statistics
.
-DP
What models are used in assembly output filling
(Pattern)
And alternative mode
.
-fpretend-float
Cross-compilation
,
Assume that the target machine and the host use the same floating point format
.
It causes floating point constants for output errors
,
But when running on the target machine
,
Real instruction sequences are likely to
GNU CC
Hopeful
.
-SAVE-TEMPS
Save those usually
``
temporary
'' '
Intermediate file
;
Place in the current directory
,
And naming according to the source file
.
therefore
,
use
`-c -save-temps'
Option compilation
`foo.c '
Generate
`foo.cpp '
with
`foo.s'
as well as
`foo.o '
file
.
-Print-file-name = library
Display library file
Library
Full path name
,
Use this library when connecting
---
What else is not
.
According to this option
, GNU CC
Neither compilation
,
Nor connected
,
Only display the file name
.
-Print-libgcc-file-name
with
`-Print-file-name = libgcc.a '
same
.
-Print-prog-name = program is similar to
`-Print-file-name ',
But find the program
PROGRAM
Such as
`CPP '.
Optimization option (Optimization Option)
These options control a variety of optimization measures
:
-O
-O1
optimization
.
For large functions
,
Optimize compilation to occupy a slight time and quite large memory
.
When not using the `-o 'option, the compiler's goal is to reduce the compilation overhead, so that the compile results can be debugged. Statement is independent: If you use a breakpoint abort from two statements, you can re-assign any variables. Or point the program counter to other statements in the function body, and accurately obtain the results you expect from the source program.
When you don't use the `-o 'option, you only declare that the REGISTER variable is allocated to use the register. The compile result is slightly inferior to the PCC that does not need to be` -o' option.
Using the `-o 'option, the compiler will try to reduce the size and execution time of the target code.
If you specify the `-o 'option,` -fthread-jumps' option will be opened. On the machine with DELAY SLOT, `-fdelayed-branch' option will be opened. Even if there is no frame Frame Pointer also supports debugging machines, `-fomit-frame-pointer 'option will be opened. Some machines may also open other options.
-O2
More optimization
.
In addition to optimization options involving spatial and speed exchange
,
Execute almost all optimization
.
For example, do not loop
(loop unrolling)
And functions
Inline.
with
-O
Option comparison
,
This option adds both compilation time
,
Also improve the operation of generating code
.
-O3
More more
.
Except
-O2
All
,
It is also open
-finline-functions
Option
.
-O0
Non-optimization
.
If you specify multiple -O options, no matter how the belt does not have a number, the last option is an option to take effect.
Options such as `-fflag 'describe some of the machine-independent switches. Most switches have affirmative and negation two formats; the negative format of` -ffoo' switch option should be `-fno-foo '. The following list is only displayed A format --- that is not a default option. You can construct another format by removing or adding `NO- '.
-ffloat-store
Do not store floating point variables in registers
.
This prevents unwanted high precision on some machines
,
Such as
68000
Floating point register
(
From
68881)
The precision saved exceeded
Double
Accuracy should have
.
For most programs, only high precision is only good. However, some procedures are strictly dependent on the definition of the IEEE floating point number. You can use the `-ffloat-store 'option for such programs.
-fmemoize-lookups
-fsave-memoized
Exploration method
(HEURISTIC)
Make faster compilation
(
Only
C ).
Discovery is not used by default
.
Since the exploration method is only valid for some input files
,
Compiling speeds of other programs will become slower
.
The first compiler must establish a call to the member function (or reference to member data). It must (1) to determine if this class implements the member function of that name; (2) Decide which member function of the call (involved) Specifies which type of conversion you need to do); (3) Check if the member function is visible to the caller. All of these constitute a slower compilation. General circumstances, the second to member functions (or reference to member data), It must be processed again after the same length. This means like this code
COUT << "this" << p << "HAS" << n << "leggs./n";
It is necessary to do six traverses for the entire three steps. By using software cache, `` hit '' can significantly reduce this price. However, unfortunately, using this kind of cache must implement other mechanisms, bringing its own overhead. `-fmemoize-lookups 'option Open the software cache. Because the function of the function is different, the access to the member and member function may be different, G may need to refresh the cache. Use the` -fmemoize-lookups' option, Each compilation is completely refreshed. The `-fsave-memoized 'option also enables the same cache, but when the compiler finds the access rights generated by the final compiled function, the next function is the same, the compiler Keep the cache content. This is very useful when you define a number of member functions for a class: In addition to some other classes, each member function has exactly the same access as the other member functions, so there is no need to refresh the cache.
-fno-default-inline
Silently do not embed the member function
,
Because they define within the scope of the class
(
only
C ).
-fno-defer-pop
Once the function returns
,
The parameter is immediately popped up
.
Machines that must be popped up after the call function
,
The compiler generally allows the parameters that call several functions to stack on the stack
,
Then pop up all times
.
-FFORCE-MEM
Before making mathematical operations, send the memory operand to use into the register
.
Convert memory to a potential public sub-expression
,
It may produce a better target code
.
If they are not public sub-expressions
,
Instruction combinations should eliminate their respective registers
.
I am happy to listen to different opinions.
.
-FFORCE-ADDR
Before making mathematical operations, send memory address constants to use into the register
.
It may be
`-fforce-mem '
A good target code
.
I am happy to listen to different opinions.
.
-FOMIT-FRAME-POINTER
For no frame pointer
Frame Pointer
The function
,
Do not save the frame pointer in the register
.
This avoids preservation
,
Setting and recovering the instruction of the frame pointer
;
Also provide an additional register for many functions
.
But in most machines will not be debugged
.
Some machines, such as VAX, this option is invalid because the standard call sequence automatically processes the frame pointer, and does not save anything by pretending to exist without saving anything. Machine Description Macro Frame_Pointer_Required Controls whether the target is supported.
-finline-functions
Integrate all simple functions into invigoors
.
The compiler explores what functions are simple enough
,
Worth this integration
.
If all the calls of all given functions are integrated, and the function declares that it is static, then the GCC has the right to not follow the assembly code output function.
-fcaller-saves
Allows value in the register
,
But this scheme is usually used by each function called
,
therefore
GCC
Generate additional code
,
Save and restore register content before and after the function call
.
Such assignments are achieved only when the generated code looks better than the anti-the results.
.
Some machines This option defaults to allow, usually these machines do not call the protection register instead of use.
-fkeep-inline-functions
Even if all calls are integrated
,
And the function is declared
STATIC,
Still output this function a separate
,
Releaseable version
.
-fno-function-cse
Do not store the function address into the register
;
Let the instruction of the call fixed function explicitly give the function address
.
This option generates a low efficiency target code, but if this option is not used, some unusual HACK, change the output of the handup, may be confused due to optimization.
-fno-peephole
Prohibit any machine
Peephole
optimization
.
-FFAST-MATH
This option is for speed optimization
,
allow
GCC
Violate certain
ANSI
or
IEEE
rule
/
specification
.
E.g
,
It allows compiler to assume
SQRT
The parameters of the function are non-negative numbers
.
This option is not opened by any `-o 'option, because for mathematics functions implemented strictly on the IEEE or ANSI rules / specification, the program may generate errors. The following options control a specific optimization.` -O2' option open Most optimization items, in addition to `-funroll-loops 'and` -funroll-all-loops' items.
The `-o 'option typically opens the` -fthread-jumps' and `-fdelayed-branch' optimization item, but the default optimization item on a particular machine is possible to change.
If you are very needed in special case, you can use the following options.
-fstregth-reduuce
Reduce loop strength
(Loop Stregth Reduction)
optimization
,
And eliminate repeating variables
.
-fthread-jumps
The location where the optimization is
,
If there is another condition of the destination of a jump branch
,
And this condition is included within the previous comparative statement
,
So implement optimization
.
According to the conditions
True
or
False,
The front branch redirects to the second branch or followed behind the second branch
.
-funroll-loops
Perform loop
(loop unrolling)
optimization
.
Only the number of cycles can be implemented at the time of compile or when the runtime is compiled.
.
-funroll-all-loops
Perform loop
(loop unrolling)
optimization
.
Implement all loops
.
Usually make the program run slower
.
-fcse-follow-jumps
In public sub-expression
(Common SubExpression Elimination)
when
,
If there is no other path to reach a destination of a jump
,
Sweake this
JUMP
instruction
.
E.g
,
in case
CSE
Encounter
Else
Squeaky
IF
Statement
,
When the condition is tested
False
Time
CSE
Follow
JUMP
Behind
.
-fcse-skip-blocks
It is similar to
`-fcse-follow-jumps'
Option
,
but
CSE
Back behind the condition
,
Condition jump skipping statement block
(Block).
in case
CSE
A simple
IF
Statement
,
Without
Else
Clause
, `-fcse-skip-blocks'
Option will result
CSE
follow
IF
Generated jump behind
.
-freun-cse-after-loop
After performing loop optimization
,
Return public sub-expression
.
-felide-constructionors
If you look reasonably, you will omit the omnifier.
(
only
C ).
According to this option
,
For the following code
, GNU C
Directly call
foo
initialization
Y,
No need to pass temporary variables
:
A foo (); a y = foo ();
If there is no such option, GNU C first initializes Y, then assigns Y; then the result of the FOO is assigned to the temporary variable; finally, replace the initial value of `Y 'with a temporary variable.
The ANSI C standard specifies the default behavior (`-fno-elide-constructors '). If the program is constructed,` -felide-constructors' option can make the program different performance, because some constructs may be ignored transfer.
-FEXPENSIVE-OPTIMIZATIONS
Execute some relative overhead a secondary optimization
.
-FDELAYED-BRANCH
If you support the target machine
,
It tries to rearrange the instruction
,
To use delay branch
(Delayed Branch)
Instruction void behind the instruction
.
-fschedule-insns
If you support the target machine
,
It tries to rearrange the instruction
,
In order to eliminate the execution pause caused by data
.
This can help floating point operations or memory access slower machines
,
Allow other instructions to be executed
,
Until the instruction or floating point operation is completed
.
-fschedule-insns2
Similar
`-fschedule-insns'
Option
,
But after the register is allocated
,
Need an additional instruction scheduling process
.
Relatively fewer registers
,
Moreover, the memory command is greater than one cycle machine.
,
This option is especially useful
.
Target Options (Target Option)
Default
, GNU CC
Compile the target code of this type
.
However, you can also install him into a cross compiler.
,
Compiler for other models
.
In fact
,
Different target machines
,
Can be installed at the same time
GNU CC
Corresponding configuration
.
Then
`-b '
Option Specify the target machine
.
By the way, the new version and the old version of the GNU CC can coexist. One version (which may be the latest) is the default version, but sometimes you want to use other versions.
-B Machine
parameter
Machine
Point out the compiled target model
.
This option is used to install the cross-compiler
GNU CC.
The value of the parameter Machine is the same as the machine type set when the GNU CC cross-compiler is configured. For example, if the cross-compiler is configured with `Configure i386v ', meaning compiling the System V target code on 80386, then you can pass` -b i386v 'Run the cross compiler.
If you do not specify a `-b 'option, you usually refer to compiling the native target code.
-V Version
parameter
Version
Point out which version is running
GNU CC.
This option is used to install multiple versions
GCC.
E.g
,
in case
Version
Yes
`2.0 ',
Mean operation
GNU CC 2.0
Version
.
If there is no designated `-v 'option, the default version depends on the installation of GNU CC, which is generally recommended to use universal versions.
Machine Related Option (Machine Dependent Option)
Every target model has its own special options
,
These options are used
`-m '
Switch guidance
,
Select different hardware models or configuration
---
E.g
, 68010
still is
68020,
Is there a floating point association processor
.
By specifying options
,
One version of the installation compiler can compile all models or configurations
.
In addition, some configuration of the compiler supports additional special options, usually in order to compatibility with this platform on the command line.
Below is a `-m 'option for 68000 series:
-m68000
-mc68000
Output
68000
Target code
.
If the compiler is based on
68000
System configuration
,
This option is the default option
.
-M68020
-mc68020
Output
68020
Target code
(
Instead of
68000).
If the compiler is based on
68020
System configuration
,
This option is the default option
.
-M68881
Output included
68881
Target code for floating point instructions
.
For most
68020
System This is the default option
,
Specified unless the compiler is set
-NFP.
-M68030
Output
68030
Target code
.
If the compiler is based on
68030
System configuration
,
This option is the default option
.
-m68040
Output
68040
Target code
.
If the compiler is based on
68040
System configuration
,
This option is the default option
.
-M68020-40
Output
68040
Target code
,
But do not use new instructions
.
The generated code can be
68020/68881
on
,
Can also
68030
or
68040
Upper effect
.
-MFPA
Output included
Sun FPA
Target code for floating point instructions
.
-MSoft-float
The output contains a target code that is called floating point library.
.
caveat:
The reserved library is not
GNU CC
Part
.
Generally say
GCC
Use this model
C
Compiler
,
But when you cross-compile, you can't use it directly.
.
You must manage your own function library for cross-compilation
.
-mshort
think
int
Type is
16
Bit wide
,
Equivalent
Short Int.
-Mnobitfield
Do not use the domain
(Bit-Field)
instruction
. `-m68000 '
Implicit specified
`-mnobitfield '.
-Mbitfield
Use bitmap
. `-m68020 '
Implicit specified
`-mbitfield '.
If you use unmarkable
GCC,
This is the default option
.
-mrtd
Adopt another function call
,
Function accepts fixed number of parameters
,
use
RTD
Command returns
,
This command returns the parameters in the stack when the instruction is returned.
.
This method enables the caller to save one instruction
,
Because he does not need to pop up parameters
.
This call agreed is not compatible with UNIX normal calls. So if you need to call the library function compiled by the UNIX compiler, you cannot use this option.
In addition, all parameters variable functions must provide a function prototype (including Printf); otherwise the compiler generates an error call code.
Also, if you carrier too much parameters while calling a function, the compiler will generate a serious error code. (Under normal circumstances, excess parameters are ignored safely.)
68010 and 68020 processors support RTD instructions, but 68000 does not support.
Here is the `-m 'option for VAX definitions:
-munix
No output some jump instructions
(AOBLEQ
and many more
), VAX
of
Unix
Assembler cannot span long range
(Long Ranges)
Process
.
-MGNU
If used
Gnu
Assembler
,
Output those jump instructions
,
-mg
Output
G-Format
Floating point number
,
replace
D-Format.
Here is the `-m 'option switch supported by SPARC:
-MFPU
-mhard-float
Output a target code containing floating point instructions
.
This is the default option
.
-Mno-fpu
-MSoft-float
The output contains a target code that is called floating point library.
.
caveat:
Not
Sparc
provide
Gnu
Floating point library
.
Generally speaking, using this model
C
Compiler
,
But you can't use it directly for cross-compilation
.
You must arrange yourself
,
Provide libraries for cross-compilation
.
-msoft-float changes the call agreement in the output file; therefore only use this option to compile the entire program.
-Mno-Epilogue
-mepilogue
use
-mepilogue
default
)
Option time
,
The compiler always places the exit code of the function in the tail of the function.
.
Any exit statement in the middle of a function
(
E.g
C
middle
Return
Statement
)
The jump instruction will result in the end of the function
.
When using the -Mno-epilogue option, the compiler is as used to extract the exit code at each function exit point.
-MNO-V8
-MV8
-MspARCLITE
These three options choose different kinds of
Sparc
system
.
By default (unless specifically for Fujitsu SparClite configuration), the GCC generates the SPARC V7 target code.
-MV8 generates a SPARC V8 target code. The only difference between his and V7 target code is that the compiler generates an integer multiplication and integer division instructions, and SPARC V8 supports the instruction, and the V7 system is not supported.
-msparclite generates a SparClite target code. Added an integer multiplication supported by SparClite, an Integer Divide STEP AND SCAN (FFS) instruction. The V7 system does not support these instructions.
-MCyPress
-MSUPERSPARC
These two options selection processor model
,
Code optimization for the processor
.
-mcypress option (default) Enables the compiler to optimize the CYPRESS CY7C602 chip, the SparcStation / SparcServer 3xx series uses this chip. This option also applies to old SparcStation 1, 2, IPX and other models ..- MSUPARC option makes compilation The SUPERSPARC processor optimizes code, SparcStation 10, 1000, and 2000 Series use this chip. At the same time, the option enables a complete SPARC V8 instruction set.
Here is the `-m 'option for Convex definitions:
-mc1
Output
C1
Target code
.
Compiler
C1
Moderate
,
This is the default option
.
-mc2
Output
C2
Target code
.
Compiler
C2
Moderate
,
This is the default option
.
-Margcount
Place a parameter count in front of each parameter list
Argument Count Word.
Some irreplaceable
Convex
with
VAX
Program requires this parameter number
(
The debugger doesn't need him
,
Unless the function comes with a list of viable parameters
;
This information is stored in the symbol table
.)
-Mnoargcount
Ignore parameter counting
.
If you use unmarkable
GCC,
This is the default option
.
Below is the `-m 'option defined for the AMD AM29000:
-mdw
The generated target code is considered
DW
Position
,
that is
,
Bytes and half-word operations are supported by hardware
.
This option is the default option
.
-MNODW
The generated target code is considered
DW
Unsettled
.
-MBW
The generated target is considered to support bytes and half-word write operations.
.
This option is the default option
.
-MNBW
The generated target code considers that the system does not support bytes and half-word write operations
.
This option is hidden.
`-mnodw '
Option
.
-msmall
Use small memory mode
,
Small memory mode assumes that all functions of all functions are located in one
256 kB
Segment
,
Or all the absolute addresses are less than
256K.
This can be used
Call
Instruction
Const, Consth, Calli
Instruction sequence
.
-mlarge
Assume that you can't use
Call
instruction
;
This is the default option
.
-M29050
Output
AM29050
Target code
.
-M29000
Output
AM29000
Target code
.
This is the default option
.
-Mkernel-registers
Generated target code reference
GR64-GR95
Register instead of
GR96-GR127
register
.
This option can be used to compile kernel code
,
The kernel needs a set of global registers
,
These global registers and registers used by user mode are completely unrelated
.
Note that when using this option, the register name in the `-f 'option must be Normal, User-Mode, Names.
-muser-registers
Use ordinary global register set
GR96-GR127.
This is the default option
.
-MSTACK-CHECK
Insert one after each stack adjustment
__MSP_CHECK
transfer
.
This option is often used for kernel code
.
Here is the `-m 'option defined for the Motorola 88K system:
-m88000
The generated target code can be
M88100
with
M88110
Normal work
.
-M88100
The generated target code is
M88100
The best work
,
But
M88110
Run on
.
-m88110
The generated target code is
M88110
The best work
,
May not be in
M88100
Run on
.
-midentify-revision
In the output of the assembler contains one
Ident
instruction
,
Record source file name
,
Compiler name and version
,
Time standard
,
And useful compilation options
,
-Mno-underscores
Output of the assembler
,
No next line does not add a next line in front of the symbol name
.
The default is to increase the underscore prefix in front of each name
.
-Mno-check-zero-division-mcheck-zero-division
Early model
88K
System in addition to zero operation
,
Many machines cannot be saved in a specific situation
.
Use these options to avoid inclusion
(
Or you can clearly include
)
Additional code
,
These codes can check
,
Send exception signal
GCC
all
88K
Configuration default
`-mcheck-zero-division '
Option
.
-MOCS-Debug-Info
-Mno-OCS-Debug-Info
contain
(
Ignore
)
Additional debugging information
(
About the use of registers in each stack structure
88Open Object Compatibility Standard, `` OCS '',
Due to this information
GDB
No need for these extra information
DG / UX, SVR4,
with
Delta 88 SVR3.2
The default configuration is to include debugging information
,
other
88K
The default configuration of the model is to ignore this information
.
-MOCS-FRAME-POSITION
-Mno-OCS-Frame-Position
Mandatory
(
Not required
)
Store register values to the specified location in the stack frame structure
(
press
OCS
instruction of
DG / UX, Delta88 SVR3.2
with
BCS
Default configuration
`-mocs-frame-position '
Option
;
other
88K
The default configuration of the model is
`-mno-ivs-frame-position '.
-Moptimize-arg-area
-Mno-Optimize-Arg-area
Control how to store function parameters in a stack structure
. `-moptimize-arg-area '
save space
,
But it is likely to drop some debugger
(
Not
GDB). `-Mno-Optimize-arg-area '
Calve better than standard options
.
by default
GCC
Non-optimization parameter domain
.
-mshort-data-
Num
Pass and
r0
Association
,
Send smaller data reference
(Data Reference),
This allows you to transfer a value with a single instruction.
(
Not a normal dual directive
).
Users in the options
Num
Which data reference control changes?
.
E.g
,
If you specify
`-mshort-data-512 ',
Then the affected data is less than
512
Byte data movement
. -MSHORT-DATA-NUM
Option pair is greater than
64K
of
Num
invalid
.
-Mserialize-Volatile
-Mno-Serialize-Volatile
produce
,
Or do not generate code to ensure consistent consistency of volatility
.
For commonly used processor sub-models, the GNU CC always guarantees such consistency by default. How to implement the results consistently depend on the processor sub-model.
The M88100 processor does not rearrange the memory reference, so the access results are always consistent. If you are using the `-m88100 'option, the GNU CC does not produce special instructions consistent with the results.
The memory reference order of the M88110 processor does not always comply with the order of the instruction request. In particular, a read instruction may be executed before the previous storage instruction. Under the multiprocessor environment, the charter access to disturb the variable memory access Consistency. Therefore, when using `-m88000 'or` -m88110' option, the GNU CC produces special instructions when appropriate, forcing the execution order correct.
These additional code used to ensure consistency may affect the performance of the program. If you confirm that you can give up this guarantee, you can use the `-Mno-serialize-volatile 'option.
If you use a `-m88100 'option, you need to match the results of the M88110 processor, you should add the` -mserialize-volatile' option.
