Next: Constraints, Previous: Inline, Up: C Extensions
In an assembler instruction using asm, you can specify the
operands of the instruction using C expressions. This means you need not
guess which registers or memory locations will contain the data you want
to use.
You must specify an assembler instruction template much like what appears in a machine description, plus an operand constraint string for each operand.
For example, here is how to use the 68881's fsinx instruction:
asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
Here angle is the C expression for the input operand while
result is that of the output operand. Each has `"f"' as its
operand constraint, saying that a floating point register is required.
The `=' in `=f' indicates that the operand is an output; all
output operands' constraints must use `='. The constraints use the
same language used in the machine description (see Constraints).
Each operand is described by an operand-constraint string followed by the C expression in parentheses. A colon separates the assembler template from the first output operand and another separates the last output operand from the first input, if any. Commas separate the operands within each group. The total number of operands is currently limited to 30; this limitation may be lifted in some future version of GCC.
If there are no output operands but there are input operands, you must place two consecutive colons surrounding the place where the output operands would go.
As of GCC version 3.1, it is also possible to specify input and output
operands using symbolic names which can be referenced within the
assembler code. These names are specified inside square brackets
preceding the constraint string, and can be referenced inside the
assembler code using %[name] instead of a percentage sign
followed by the operand number. Using named operands the above example
could look like:
asm ("fsinx %[angle],%[output]"
: [output] "=f" (result)
: [angle] "f" (angle));
Note that the symbolic operand names have no relation whatsoever to other C identifiers. You may use any name you like, even those of existing C symbols, but you must ensure that no two operands within the same assembler construct use the same symbolic name.
Output operand expressions must be lvalues; the compiler can check this.
The input operands need not be lvalues. The compiler cannot check
whether the operands have data types that are reasonable for the
instruction being executed. It does not parse the assembler instruction
template and does not know what it means or even whether it is valid
assembler input. The extended asm feature is most often used for
machine instructions the compiler itself does not know exist. If
the output expression cannot be directly addressed (for example, it is a
bit-field), your constraint must allow a register. In that case, GCC
will use the register as the output of the asm, and then store
that register into the output.
The ordinary output operands must be write-only; GCC will assume that the values in these operands before the instruction are dead and need not be generated. Extended asm supports input-output or read-write operands. Use the constraint character `+' to indicate such an operand and list it with the output operands. You should only use read-write operands when the constraints for the operand (or the operand in which only some of the bits are to be changed) allow a register.
You may, as an alternative, logically split its function into two
separate operands, one input operand and one write-only output
operand. The connection between them is expressed by constraints
which say they need to be in the same location when the instruction
executes. You can use the same C expression for both operands, or
different expressions. For example, here we write the (fictitious)
`combine' instruction with bar as its read-only source
operand and foo as its read-write destination:
asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
The constraint `"0"' for operand 1 says that it must occupy the same location as operand 0. A number in constraint is allowed only in an input operand and it must refer to an output operand.
Only a number in the constraint can guarantee that one operand will be in
the same place as another. The mere fact that foo is the value
of both operands is not enough to guarantee that they will be in the
same place in the generated assembler code. The following would not
work reliably:
asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
Various optimizations or reloading could cause operands 0 and 1 to be in
different registers; GCC knows no reason not to do so. For example, the
compiler might find a copy of the value of foo in one register and
use it for operand 1, but generate the output operand 0 in a different
register (copying it afterward to foo's own address). Of course,
since the register for operand 1 is not even mentioned in the assembler
code, the result will not work, but GCC can't tell that.
As of GCC version 3.1, one may write [name] instead of
the operand number for a matching constraint. For example:
asm ("cmoveq %1,%2,%[result]"
: [result] "=r"(result)
: "r" (test), "r"(new), "[result]"(old));
Sometimes you need to make an asm operand be a specific register,
but there's no matching constraint letter for that register by
itself. To force the operand into that register, use a local variable
for the operand and specify the register in the variable declaration.
See Explicit Reg Vars. Then for the asm operand, use any
register constraint letter that matches the register:
register int *p1 asm ("r0") = ...;
register int *p2 asm ("r1") = ...;
register int *result asm ("r0");
asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
In the above example, beware that a register that is call-clobbered by
the target ABI will be overwritten by any function call in the
assignment, including library calls for arithmetic operators.
Also a register may be clobbered when generating some operations,
like variable shift, memory copy or memory move on x86.
Assuming it is a call-clobbered register, this may happen to r0
above by the assignment to p2. If you have to use such a
register, use temporary variables for expressions between the register
assignment and use:
int t1 = ...;
register int *p1 asm ("r0") = ...;
register int *p2 asm ("r1") = t1;
register int *result asm ("r0");
asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
Some instructions clobber specific hard registers. To describe this, write a third colon after the input operands, followed by the names of the clobbered hard registers (given as strings). Here is a realistic example for the VAX:
asm volatile ("movc3 %0,%1,%2"
: /* no outputs */
: "g" (from), "g" (to), "g" (count)
: "r0", "r1", "r2", "r3", "r4", "r5");
You may not write a clobber description in a way that overlaps with an
input or output operand. For example, you may not have an operand
describing a register class with one member if you mention that register
in the clobber list. Variables declared to live in specific registers
(see Explicit Reg Vars), and used as asm input or output operands must
have no part mentioned in the clobber description.
There is no way for you to specify that an input
operand is modified without also specifying it as an output
operand. Note that if all the output operands you specify are for this
purpose (and hence unused), you will then also need to specify
volatile for the asm construct, as described below, to
prevent GCC from deleting the asm statement as unused.
If you refer to a particular hardware register from the assembler code, you will probably have to list the register after the third colon to tell the compiler the register's value is modified. In some assemblers, the register names begin with `%'; to produce one `%' in the assembler code, you must write `%%' in the input.
If your assembler instruction can alter the condition code register, add `cc' to the list of clobbered registers. GCC on some machines represents the condition codes as a specific hardware register; `cc' serves to name this register. On other machines, the condition code is handled differently, and specifying `cc' has no effect. But it is valid no matter what the machine.
If your assembler instructions access memory in an unpredictable
fashion, add `memory' to the list of clobbered registers. This
will cause GCC to not keep memory values cached in registers across the
assembler instruction and not optimize stores or loads to that memory.
You will also want to add the volatile keyword if the memory
affected is not listed in the inputs or outputs of the asm, as
the `memory' clobber does not count as a side-effect of the
asm. If you know how large the accessed memory is, you can add
it as input or output but if this is not known, you should add
`memory'. As an example, if you access ten bytes of a string, you
can use a memory input like:
{"m"( ({ struct { char x[10]; } *p = (void *)ptr ; *p; }) )}.
Note that in the following example the memory input is necessary,
otherwise GCC might optimize the store to x away:
int foo ()
{
int x = 42;
int *y = &x;
int result;
asm ("magic stuff accessing an 'int' pointed to by '%1'"
"=&d" (r) : "a" (y), "m" (*y));
return result;
}
You can put multiple assembler instructions together in a single
asm template, separated by the characters normally used in assembly
code for the system. A combination that works in most places is a newline
to break the line, plus a tab character to move to the instruction field
(written as `\n\t'). Sometimes semicolons can be used, if the
assembler allows semicolons as a line-breaking character. Note that some
assembler dialects use semicolons to start a comment.
The input operands are guaranteed not to use any of the clobbered
registers, and neither will the output operands' addresses, so you can
read and write the clobbered registers as many times as you like. Here
is an example of multiple instructions in a template; it assumes the
subroutine _foo accepts arguments in registers 9 and 10:
asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
: /* no outputs */
: "g" (from), "g" (to)
: "r9", "r10");
Unless an output operand has the `&' constraint modifier, GCC may allocate it in the same register as an unrelated input operand, on the assumption the inputs are consumed before the outputs are produced. This assumption may be false if the assembler code actually consists of more than one instruction. In such a case, use `&' for each output operand that may not overlap an input. See Modifiers.
If you want to test the condition code produced by an assembler
instruction, you must include a branch and a label in the asm
construct, as follows:
asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
: "g" (result)
: "g" (input));
This assumes your assembler supports local labels, as the GNU assembler and most Unix assemblers do.
Speaking of labels, jumps from one asm to another are not
supported. The compiler's optimizers do not know about these jumps, and
therefore they cannot take account of them when deciding how to
optimize. See Extended asm with goto.
Usually the most convenient way to use these asm instructions is to
encapsulate them in macros that look like functions. For example,
#define sin(x) \
({ double __value, __arg = (x); \
asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
__value; })
Here the variable __arg is used to make sure that the instruction
operates on a proper double value, and to accept only those
arguments x which can convert automatically to a double.
Another way to make sure the instruction operates on the correct data
type is to use a cast in the asm. This is different from using a
variable __arg in that it converts more different types. For
example, if the desired type were int, casting the argument to
int would accept a pointer with no complaint, while assigning the
argument to an int variable named __arg would warn about
using a pointer unless the caller explicitly casts it.
If an asm has output operands, GCC assumes for optimization
purposes the instruction has no side effects except to change the output
operands. This does not mean instructions with a side effect cannot be
used, but you must be careful, because the compiler may eliminate them
if the output operands aren't used, or move them out of loops, or
replace two with one if they constitute a common subexpression. Also,
if your instruction does have a side effect on a variable that otherwise
appears not to change, the old value of the variable may be reused later
if it happens to be found in a register.
You can prevent an asm instruction from being deleted
by writing the keyword volatile after
the asm. For example:
#define get_and_set_priority(new) \
({ int __old; \
asm volatile ("get_and_set_priority %0, %1" \
: "=g" (__old) : "g" (new)); \
__old; })
The volatile keyword indicates that the instruction has
important side-effects. GCC will not delete a volatile asm if
it is reachable. (The instruction can still be deleted if GCC can
prove that control-flow will never reach the location of the
instruction.) Note that even a volatile asm instruction
can be moved relative to other code, including across jump
instructions. For example, on many targets there is a system
register which can be set to control the rounding mode of
floating point operations. You might try
setting it with a volatile asm, like this PowerPC example:
asm volatile("mtfsf 255,%0" : : "f" (fpenv));
sum = x + y;
This will not work reliably, as the compiler may move the addition back
before the volatile asm. To make it work you need to add an
artificial dependency to the asm referencing a variable in the code
you don't want moved, for example:
asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
sum = x + y;
Similarly, you can't expect a
sequence of volatile asm instructions to remain perfectly
consecutive. If you want consecutive output, use a single asm.
Also, GCC will perform some optimizations across a volatile asm
instruction; GCC does not “forget everything” when it encounters
a volatile asm instruction the way some other compilers do.
An asm instruction without any output operands will be treated
identically to a volatile asm instruction.
It is a natural idea to look for a way to give access to the condition code left by the assembler instruction. However, when we attempted to implement this, we found no way to make it work reliably. The problem is that output operands might need reloading, which would result in additional following “store” instructions. On most machines, these instructions would alter the condition code before there was time to test it. This problem doesn't arise for ordinary “test” and “compare” instructions because they don't have any output operands.
For reasons similar to those described above, it is not possible to give an assembler instruction access to the condition code left by previous instructions.
As of GCC version 4.5, asm goto may be used to have the assembly
jump to one or more C labels. In this form, a fifth section after the
clobber list contains a list of all C labels to which the assembly may jump.
Each label operand is implicitly self-named. The asm is also assumed
to fall through to the next statement.
This form of asm is restricted to not have outputs. This is due
to a internal restriction in the compiler that control transfer instructions
cannot have outputs. This restriction on asm goto may be lifted
in some future version of the compiler. In the mean time, asm goto
may include a memory clobber, and so leave outputs in memory.
int frob(int x)
{
int y;
asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
: : "r"(x), "r"(&y) : "r5", "memory" : error);
return y;
error:
return -1;
}
In this (inefficient) example, the frob instruction sets the
carry bit to indicate an error. The jc instruction detects
this and branches to the error label. Finally, the output
of the frob instruction (%r5) is stored into the memory
for variable y, which is later read by the return statement.
void doit(void)
{
int i = 0;
asm goto ("mfsr %%r1, 123; jmp %%r1;"
".pushsection doit_table;"
".long %l0, %l1, %l2, %l3;"
".popsection"
: : : "r1" : label1, label2, label3, label4);
__builtin_unreachable ();
label1:
f1();
return;
label2:
f2();
return;
label3:
i = 1;
label4:
f3(i);
}
In this (also inefficient) example, the mfsr instruction reads
an address from some out-of-band machine register, and the following
jmp instruction branches to that address. The address read by
the mfsr instruction is assumed to have been previously set via
some application-specific mechanism to be one of the four values stored
in the doit_table section. Finally, the asm is followed
by a call to __builtin_unreachable to indicate that the asm
does not in fact fall through.
#define TRACE1(NUM) \
do { \
asm goto ("0: nop;" \
".pushsection trace_table;" \
".long 0b, %l0;" \
".popsection" \
: : : : trace#NUM); \
if (0) { trace#NUM: trace(); } \
} while (0)
#define TRACE TRACE1(__COUNTER__)
In this example (which in fact inspired the asm goto feature)
we want on rare occasions to call the trace function; on other
occasions we'd like to keep the overhead to the absolute minimum.
The normal code path consists of a single nop instruction.
However, we record the address of this nop together with the
address of a label that calls the trace function. This allows
the nop instruction to be patched at runtime to be an
unconditional branch to the stored label. It is assumed that an
optimizing compiler will move the labeled block out of line, to
optimize the fall through path from the asm.
If you are writing a header file that should be includable in ISO C
programs, write __asm__ instead of asm. See Alternate Keywords.
asmSome targets require that GCC track the size of each instruction used in
order to generate correct code. Because the final length of an
asm is only known by the assembler, GCC must make an estimate as
to how big it will be. The estimate is formed by counting the number of
statements in the pattern of the asm and multiplying that by the
length of the longest instruction on that processor. Statements in the
asm are identified by newline characters and whatever statement
separator characters are supported by the assembler; on most processors
this is the `;' character.
Normally, GCC's estimate is perfectly adequate to ensure that correct code is generated, but it is possible to confuse the compiler if you use pseudo instructions or assembler macros that expand into multiple real instructions or if you use assembler directives that expand to more space in the object file than would be needed for a single instruction. If this happens then the assembler will produce a diagnostic saying that a label is unreachable.
There are several rules on the usage of stack-like regs in asm_operands insns. These rules apply only to the operands that are stack-like regs:
An input reg that is implicitly popped by the asm must be explicitly clobbered, unless it is constrained to match an output operand.
All implicitly popped input regs must be closer to the top of the reg-stack than any input that is not implicitly popped.
It is possible that if an input dies in an insn, reload might use the input reg for an output reload. Consider this example:
asm ("foo" : "=t" (a) : "f" (b));
This asm says that input B is not popped by the asm, and that the asm pushes a result onto the reg-stack, i.e., the stack is one deeper after the asm than it was before. But, it is possible that reload will think that it can use the same reg for both the input and the output, if input B dies in this insn.
If any input operand uses the f constraint, all output reg
constraints must use the & earlyclobber.
The asm above would be written as
asm ("foo" : "=&t" (a) : "f" (b));
Output operands must specifically indicate which reg an output
appears in after an asm. =f is not allowed: the operand
constraints must select a class with a single reg.
Output operands must start at the top of the reg-stack: output operands may not “skip” a reg.
Here are a couple of reasonable asms to want to write. This asm takes one input, which is internally popped, and produces two outputs.
asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
This asm takes two inputs, which are popped by the fyl2xp1 opcode,
and replaces them with one output. The user must code the st(1)
clobber for reg-stack.c to know that fyl2xp1 pops both inputs.
asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");