This appendix describes built-in functions available when you compile on HP Tru64 UNIX and Linux Alpha systems. These functions allow you to access hardware and machine instructions directly.

Be sure to include the <machine/builtins.h> header file in your source program to access these built-in functions. Definitions for return types int64 and uint64 are contained in the header file ints.h .

Translation Macros

Compaq C++ supports the following translation macros for built-in functions:

_BBCCI(position, address) _BBSSI(position, address)

In-line Assembly Code---ASMs

The compiler supports in-line assembly code, commonly called ASMs on UNIX platforms.

Like builtin-functions, ASMs are implemented with a function-call syntax. But unlike built-in functions, to use ASMs you must include the <c_asm.h> header file containing prototypes for the three types of ASMs, and the #pragma intrinsic preprocessor directive.

These functions have the following format:

__int64 asm(const char *, ...); /* for integer operations, like MULQ */ float fasm(const char *, ...); /* for single precision float instructions, like MULS */ double dasm(const char *, ...); /* for double precision float instructions, like MULT */ #pragma intrinsic (asm, fasm, dasm)

const char *

The first argument to the asm , fasm , or dasm function contains the instruction(s) to be generated inline and the metalanguage that describes the interpretation of the arguments.

...

The source and destination arguments (if any) for the instruction being generated, and any other values used in the generated instructions.

These values are made available to the instructions through the normal argument passing conventions of the calling standard (the first integer argument is available in register R16).

The #pragma intrinsic directive in the <c_asm.h> header file is required when using ASMs. It notifies the compiler that:

• These functions are not user-defined functions.
• The special ASM processing should be applied to analyze at compile time the first argument and generate machine-code instructions as specified by the contents of the string.

The metalanguage for the argument references has the following form:

<metalanguage_sequence> : <register_alias> | <register_number> | <register_macro> ; <register_number> : "$" number ; <register_macro> : "%" <macro_sequence> ; <macro_sequence> : number | <register_name> | "f" number | "F" number | "r" number | "R" number ; <register_name> : /* argument registers: R16-R21 */ "a0" | "a1" | "a2" | "a3" | "a4" | "a5" /* return value: R0 or F0, depending on type */ | "v0" /* scratch registers: R1, R22-R24, R28 */ | "t0" | "t1" | "t2" | "t3" | "t4" /* save registers: R2-R15 */ | "s0" | "s1" | "s2" | "s3" | "s4" | "s5" | "s6" | "s7" | "s8" | "s7" | "s8" | "s9" | "s10" | "s11" | "s12" | "s13" /* stack pointer: R30 */ | "sp" | "SP" | "$sp" | "$SP" | "RA" | "ra" /* return addr: R26 */ | "PV" | "pv" /* procedure value: R27 */ | "AI" | "ai" /* arg info: R25 */ | "FP" | "fp" /* frame pointer: R29 */ | "RZ" | "rz" | "zero" /* sink/source: R31 == zero */

Syntactically, the metalanguage can appear anywhere within an instruction sequence.

The literal string that contains instructions, operands, and metalanguage must follow the general form:

<string_contents> : <instruction_seq> | <string_contents> ";" <instruction_seq> | error | <string_contents> error ; <instruction_seq> : instruction_operand | directive ;

An instruction_operand is generally recognized as an assembly language instruction separated by whitespace from a sequence of comma-separated operands.

You can code multiple instruction sequences into one literal string, separating them by semicolons.

Because adjacent string literals are concatenated into a single string, successive instructions can be written as separate strings, one per line (as is normally done in assembly language) as long as each instruction is terminated by a semicolon (as shown in the examples).

There are semantic and syntax rules associated with ASMs:
The first argument to an ASM call is interpreted as the instructions to be assembled in the metalanguage, and must be fully understood by the compiler at compile time. Therefore, it must be a literal string (or a macro expanding to a literal string) and must not be a run-time value containing a string. Therefore, the following are not allowed: indirections, table lookups, structure dereferences, and so on.
The remaining arguments are loaded into the argument registers like normal function arguments, except that the second argument to the ASM call is treated as the first argument for purposes of the calling standard.
For example, in the following test, the six arguments are loaded into arg registers a0 through a5, and the result of each subexpression is stored in the value return register v0. Since v0 is the calling standard's return value register (R0 for an integer function), the result of the final MULQ is the value returned by the "call":

if (asm("mulq %a0, %a1, %v0;" "mulq %a2, %v0, %v0;" "mulq %a3, %v0, %v0;" "mulq %a4, %v0, %v0;" "mulq %a5, %v0, %v0;", 1, 2, 3, 4, 5, 6) != 720){ error_cnt++; printf ("Test failed\n"); }

The following example does not work. There is no value loaded into the floating-point return register. Furthermore, it results in a compile-time warning stating that r2 is used before it is set, because the arguments are loaded into the arg registers and not into r2:

z = fasm("mulq %r2, %a1 %r5", x=10, y=5);

The correct way of doing this is to specify an argument register number in place of r2. A correct version of the above would be:

z = fasm("mulq %a0, %a1, %a1;" "stq %a1, 0(%a2);" "ldt %f0, 0(%a2);" "cvtqf %f0, %f0;", x=10, y=5, &temp);

Note that the memory location used for the transfer from integer to floating-point register is made available to the asm code by passing as an argument the address of a variable allocated in the C code for that purpose.

• A return register must be specified in the metalanguage for the result to appear in the expected place.
• For intructions that do not take any argument and do not have a return type, leave out the arguments. For example:

  asm("MB");

Privileged Architecture Library Code Instructions

The following Privileged Architecture Library Code (PALcode) instructions are available as built-in functions:

__PAL_GENTR __PAL_HALT __PAL_BPT __PAL_BUGCHK __PAL_DRAINA

__PAL_BPT

This function is provided for program debugging. It switches the processor to kernel mode and pushes registers R2 to R7, the updated PC, and PS onto the kernel stack. It then dispatches to the address in the breakpoint vector, which is stored in a control block.

This function has the following format:

void __PAL_BPT (void);

__PAL_BUGCHK

This function is provided for error reporting. It switches the processor to kernel mode and pushes registers R2 to R7, the updated PC, and PS onto the kernel stack. It then dispatches to the address in the bugcheck vector, which is stored in a control block.

This function has the following format:

void __PAL_BUGCHK (unsigned long);

__PAL_DRAINA

This function stalls instruction issuing until all prior instructions are guaranteed to complete without incurring aborts.

This function has the following format:

void __PAL_DRAINA (void);

__PAL_GENTRAP

This function is used for reporting run-time software conditions.

This function has the following format:

void __PAL_GENTRAP (uint64 encoded_software_trap);

encoded_software_trap

The particular software condition that has occurred.

__PAL_HALT

This function halts the processor when executed by a process running in kernel mode. This is a privileged function.

This function has the following format:

void __PAL_HALT (void);

Absolute Value ( __ABS)

The __ABS built-in is functionally equivalent to its counterpart, abs , in the standard header file <stdlib.h> .

Its format is also the same:

#include <stdlib.h> int __ABS (int x);

This built-in function does, however, offer performance improvements because there is less call overhead associated with its use.

If you include <stdlib.h> , the built-in function is automatically used for all occurrences of abs . To disable the built-in function, use #undef abs .

Add Aligned Word Interlocked ( __ADAWI)

The __ADAWI function adds its source operand to the destination. This function is interlocked against similar operations by other processors or devices in the system.

This function has the following format:

int __ADAWI (short src, short *dest);

src

The value to be added to the destination.

dest

A pointer to the destination. The destination must be aligned on a word boundary. (You can achieve alignment using the _align storage-class modifier.)

The __ADAWI function returns a simulated VAX processor status longword (PSL).

Add Atomic Longword ( __ADD_ATOMIC_LONG)

The __ADD_ATOMIC_LONG function adds the specified expression to the longword data segment pointed to by the address parameter within a load-locked/store-conditional code sequence.

This function has the following format:

int __ADD_ATOMIC_LONG int fnc(volatile void *, int, ...);

address

The address of the data segment.

expression

An integer expression.

...

An optional retry count of type int . If specified, the retry count indicates the number of times the operation is attempted. If the operation cannot be performed successfully in the specified number of retries, a value of 0 is returned.

A value of 1 is returned upon successful completion.

Add Atomic Quadword ( __ADD_ATOMIC_QUAD)

The __ADD_ATOMIC_QUAD function adds the specified expression to the quadword data segment pointed to by the address parameter within a load-locked/store-conditional code sequence.

This function has the following format:

int __ADD_ATOMIC_QUAD (void *address, int expression, ...);

address

The address of the data segment.

expression

An integer expression.

...

An optional retry count of type int . If specified, the retry count indicates the number of times the operation is attempted. If the operation cannot be performed successfully in the specified number of retries, a value of 0 is returned.

A value of 1 is returned upon successful completion.

AND Atomic Longword ( __AND_ATOMIC_LONG)

The __AND_ATOMIC_LONG function performs a bit-wise or arithmetic AND of the specified expression with the longword data segment pointed to by the address parameter within a load-locked/store-conditional code sequence.

This function has the following format:

int __AND_ATOMIC_LONG (void *address, int expression, ...);

address

The address of the data segment.

expression

An integer expression.

...

An optional retry count of type int . If specified, the retry count indicates the number of times the operation is attempted. If the operation cannot be performed successfully in the specified number of retries, a value of 0 is returned.

A value of 1 is returned upon successful completion.

AND Atomic Quadword ( __AND_ATOMIC_QUAD)

The __AND_ATOMIC_QUAD function performs a bit-wise or arithmetic AND of the specified expression with the aligned quadword pointed to by the address parameter within a load-locked/store-conditional code sequence.

This function has the following format:

int __AND_ATOMIC_QUAD (void *address, int expression, ...);

address

The address of the aligned quadword.

expression

An integer expression.

...

An optional retry count of type int . If specified, the retry count indicates the number of times the operation is attempted (which will be at least once, even if the count argument is 0). If the operation cannot be performed successfully in the specified number of retries, a value of 0 is returned.

A value of 1 is returned upon successful completion.

Atomic Add Longword (__ATOMIC_ADD_LONG)

The __ATOMIC_ADD_LONG function adds the specified expression to the aligned longword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the longword before the addition was performed.

This function has one of the following formats:

int __ATOMIC_ADD_LONG (volatile void *address, int expression); int __ATOMIC_ADD_LONG_RETRY (volatile void *address, int expression, int retry, int *status);

address

The longword-aligned address of the data segment.

expression

An integer expression.

retry

A retry count of type int that indicates the number of times the operation is attempted (which is at least once, even if the retry argument is 0). If the operation cannot be performed successfully in the specified number of retries, the function returns without updating the longword.

status

A pointer to an integer that is set to 0 if the operation did not succeed within the specified number of retries, and set to 1 if the operation succeeded.

Atomic Add Quadword (__ATOMIC_ADD_QUAD)

The __ATOMIC_ADD_QUAD function adds the specified expression to the aligned quadword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the quadword before the addition was performed.

This function has one of the following formats:

int __ATOMIC_ADD_QUAD (volatile void *address, int expression); int __ATOMIC_ADD_QUAD_RETRY (volatile void *address, int expression, int retry, int *status);

address

The quadword-aligned address of the data segment.

expression

An integer expression.

retry

A retry count of type int that indicates the number of times the operation is attempted (which is at least once, even if the retry argument is 0). If the operation cannot be performed successfully in the specified number of retries, the function returns without updating the quadword.

status

A pointer to an integer that is set to 0 if the operation did not succeed within the specified number of retries, and set to 1 if the operation succeeded.

Atomic AND Longword (__ATOMIC_AND_LONG)

The __ATOMIC_AND_LONG function performs a bit-wise or arithmetic AND of the specified expression with the aligned longword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the longword before the operation was performed.

This function has one of the following formats:

int __ATOMIC_AND_LONG (volatile void *address, int expression); int __ATOMIC_AND_LONG_RETRY (volatile void *address, int expression, int retry, int *status);

address

The longword-aligned address of the data segment.

expression

An integer expression.

retry

A retry count of type int that indicates the number of times the operation is attempted (which is at least once, even if the retry argument is 0). If the operation cannot be performed successfully in the specified number of retries, the function returns without updating the longword.

status

A pointer to an integer that is set to 0 if the operation did not succeed within the specified number of retries, and set to 1 if the operation succeeded.

Atomic AND Quadword (__ATOMIC_AND_QUAD)

The __ATOMIC_AND_QUAD function performs a bit-wise or arithmetic AND of the specified expression with the aligned quadword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the quadword before the operation was performed.

This function has one of the following formats:

int __ATOMIC_AND_QUAD (volatile void *address, int expression); int __ATOMIC_AND_QUAD_RETRY (volatile void *address, int expression, int retry, int *status);

address

The quadword-aligned address of the data segment.

expression

An integer expression.

retry

A retry count of type int that indicates the number of times the operation is attempted (which is at least once, even if the retry argument is 0). If the operation cannot be performed successfully in the specified number of retries, the function returns without updating the quadword.

status

A pointer to an integer that is set to 0 if the operation did not succeed within the specified number of retries, and set to 1 if the operation succeeded.

Atomic OR Longword (__ATOMIC_OR_LONG)

The __ATOMIC_OR_LONG function performs a bit-wise or arithmetic OR of the specified expression with the aligned longword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the longword before the operation was performed.

This function has one of the following formats:

int __ATOMIC_OR_LONG (volatile void *address, int expression); int __ATOMIC_OR_LONG_RETRY (volatile void *address, int expression, int retry, int *status);

address

The longword-aligned address of the data segment.

expression

An integer expression.

retry

A retry count of type int that indicates the number of times the operation is attempted (which is at least once, even if the retry argument is 0). If the operation cannot be performed successfully in the specified number of retries, the function returns without updating the longword.

status

A pointer to an integer that is set to 0 if the operation did not succeed within the specified number of retries, and set to 1 if the operation succeeded.

Atomic OR Quadword (__ATOMIC_OR_QUAD)

The __ATOMIC_OR_QUAD function performs a bit-wise or arithmetic OR of the specified expression with the aligned quadword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the quadword before the operation was performed.

This function has one of the following formats:

int __ATOMIC_OR_QUAD (volatile void *address, int expression); int __ATOMIC_OR_QUAD_RETRY (volatile void *address, int expression, int retry, int *status);

address

The quadword-aligned address of the data segment.

expression

An integer expression.

retry

A retry count of type int that indicates the number of times the operation is attempted (which is at least once, even if the retry argument is 0). If the operation cannot be performed successfully in the specified number of retries, the function returns without updating the quadword.

status

A pointer to an integer that is set to 0 if the operation did not succeed within the specified number of retries, and set to 1 if the operation succeeded.

Atomic Increment Longword (__ATOMIC_INCREMENT_LONG)

The __ATOMIC_INCREMENT_LONG function increments by 1 the aligned longword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the longword before the operation was performed.

This function has the following format:

int __ATOMIC_INCREMENT_LONG (volatile void *address);

address

The longword-aligned address of the data segment.

Atomic Increment Quadword (__ATOMIC_INCREMENT_QUAD)

The __ATOMIC_INCREMENT_QUAD function increments by 1 the aligned quadword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the quadword before the operation was performed.

This function has the following format:

int __ATOMIC_INCREMENT_QUAD (volatile void *address);

address

The quadword-aligned address of the data segment.

Atomic Decrement Longword (__ATOMIC_DECREMENT_LONG)

The __ATOMIC_DECREMENT_LONG function decrements by 1 the aligned longword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the longword before the operation was performed.

This function has the following format:

int __ATOMIC_DECREMENT_LONG (volatile void *address);

address

The longword-aligned address of the data segment.

Atomic Decrement Quadword (__ATOMIC_DECREMENT_QUAD)

The __ATOMIC_DECREMENT_QUAD function decrements by 1 the aligned quadword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the quadword before the operation was performed.

This function has the following format:

int __ATOMIC_DECREMENT_QUAD (volatile void *address);

address

The quadword-aligned address of the data segment.

Atomic Exchange Longword (__ATOMIC_EXCH_LONG)

The __ATOMIC_EXCH_LONG function stores the value of the specified expression into the aligned longword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the longword before the operation was performed.

This function has one of the following formats:

int __ATOMIC_EXCH_LONG (volatile void *address, int expression); int __ATOMIC_EXCH_LONG_RETRY (volatile void *address, int expression, int retry, int *status);

address

The longword-aligned address of the data segment.

expression

An integer expression.

retry

A retry count of type int that indicates the number of times the operation is attempted (which is at least once, even if the retry argument is 0). If the operation cannot be performed successfully in the specified number of retries, the function returns without updating the longword.

status

A pointer to an integer that is set to 0 if the operation did not succeed within the specified number of retries, and set to 1 if the operation succeeded.

Atomic Exchange Quadword (__ATOMIC_EXCH_QUAD)

The __ATOMIC_EXCH_QUAD function stores the value of the specified expression into the aligned quadword pointed to by the address parameter within a load-locked/store-conditional code sequence and returns the value of the quadword before the operation was performed.

This function has one of the following formats:

int __ATOMIC_EXCH_QUAD (volatile void *address, int expression); int __ATOMIC_EXCH_QUAD_RETRY (volatile void *address, int expression, int retry, int *status);

address

The quadword-aligned address of the data segment.

expression

An integer expression.

retry

A retry count of type int that indicates the number of times the operation is attempted (which is at least once, even if the retry argument is 0). If the operation cannot be performed successfully in the specified number of retries, the function returns without updating the quadword.

status

A pointer to an integer that is set to 0 if the operation did not succeed within the specified number of retries, and set to 1 if the operation succeeded.

Allocate Bytes from Stack ( __ALLOCA)

The __ALLOCA function allocates n bytes from the stack.

This function has the following format:

void *__ALLOCA (unsigned int n);

n

The number of bytes to be allocated.

A pointer to the allocated memory is returned.

Single-Precision, Floating-Point Arithmetic Built-in Functions

The following built-in functions provide single-precision, floating-point chopped arithmetic:

__ADDF_C	__ADDS_C	__SUBF_C	__SUBS_C
__MULF_C	__MULS_C	__DIVF_C	__DIVS_C

They have the following format:

float __op{F,S}_C (float operand1, float operand2);

Where op is one of ADD, SUB, MUL, DIV, and {F,S} represents VAX or IEEE floating-point arithmetic.

The result of the arithmetic operation is returned.

Double-Precision, Floating-Point Arithmetic Built-in Functions

The following built-in functions provide double-precision, floating-point chopped arithmetic:

__ADDG_C	__ADDT_C	__SUBG_C	__SUBT_C
__MULG_C	__MULT_C	__DIVG_C	__DIVT_C

They have the following format:

double __op{G,T}_C (double operand1, double operand2);

Where op is one of ADD, SUB, MUL, DIV, and {G,T} represents VAX or IEEE floating-point arithmetic.

The result of the arithmetic operation is returned.

Copy Sign Built-in Functions

Built-in functions are provided to copy selected portions of single- and double-precision, floating-point numbers.

These built-in functions have the following format:

float __CPYSF (float operand1, float operand2); double __CPYS (double operand1, double operand2); float __CPYSNF (float operand1, float operand2); double __CPYSN (double operand1, double operand2); float __CPYSEF (float operand1, float operand2); double __CPYSE (double operand1, double operand2);

The copy sign built-in functions (__CPYSF and __CPYS) fetch the sign bit in operand1, concatenate it with the exponent and fraction bits from operand2, and return the result.

The copy sign negate built-in functions (__CPYSNF and __CPYSN) fetch the sign bit in operand1, complement it, concatenate it with the exponent and fraction bits from operand2, and return the result.

The copy sign exponent built-in functions (__CPYSEF and __CPYSE) fetch the sign and exponent bits from operand1, concatenate them with the fraction bits from operand2, and return the result.

Compare Store Longword ( __CMP_STORE_LONG)

The __CMP_STORE_LONG function has the following format:

int __CMP_STORE_LONG (void *source, int old_value, int new_value, void *dest);

This function compares the value pointed to by source with the longword old_value. If they are equal, the longword new_value is stored into the value pointed to by dest.

The function returns 0 if the two values are unequal, and returns 1 if the two values are equal.

Compare Store Quadword ( __CMP_STORE_QUAD)

The __CMP_STORE_QUAD function has the following format:

int __CMP_STORE_QUAD (void *source, int64 old_value, int64 new_value, void *dest);

This function compares the value pointed to by source with the quadword old_value. If they are equal, the quadword new_value is stored into the value pointed to by dest.

The function returns 0 if the two values are unequal, and returns 1 if the two values are equal.

Cosine ( __COS)

The __COS built-in function is functionally equivalent to its counterpart, cos , in the standard header file <math.h> .

Its format is also the same:

#include <math.h> double __COS (double x);

x

A radian value.

This built-in function does, however, offer performance improvements because there is less call overhead associated with its use.

If you include <math.h> , the built-in function is automatically used for all occurrences of cos . To disable the built-in function, use #undef cos .

Convert G_Floating to F_Floating Chopped ( __CVTGF_C)

The __CVTGF_C function converts a double-precision, VAX G_floating-point number to a single-precision, VAX F_floating-point number. This conversion chops to single-precision; then the 8-bit exponent range is checked for overflow or underflow.

This function has the following format:

float __CVTGF_C (double operand);

operand

A double-precision, VAX floating-point number.

Convert G-Floating to Quadword ( __CVTGQ)

The __CVTGQ function rounds a double-precision, VAX floating-point number to a 64-bit integer value and returns the result.

This function has the following format:

int64 __CVTGQ (double operand);

operand

A double-precision, VAX floating-point number.

Convert IEEE T_Floating to IEEE S_Floating Chopped ( __CVTTS_C)

The __CVTTS_C function converts a double-precision, IEEE T_floating-point number to a single-precision, IEEE S_floating-point number. This conversion chops to single-precision; then the 8-bit exponent range is checked for overflow or underflow.

This function has the following format:

float __CVTTS_C (double operand);

operand

A double-precision, IEEE floating-point number.

Convert IEEE T-Floating to Quadword ( __CVTTQ)

The __CVTTQ function rounds a double-precision, IEEE-floating-point number to a 64-bit integer value and returns the result.

This function has the following format:

int64 __CVTTQ (double operand);

operand

A double-precision, IEEE T-floating-point number.

Floating-Point Absolute Value ( __FABS)

The __FABS built-in function is functionally equivalent to its counterpart, fabs , in the standard header file <math.h> .

Its format is also the same:

#include <math.h> double __FABS (double x);

x

A floating-point number.

This built-in function does, however, offer performance improvements because there is no call overhead associated with its use.

If you include <math.h> , the built-in function is automatically used for all occurrences of fab . To disable the built-in function, use #undef fab .

Longword Absolute Value ( __LABS)

The __LABS built-in function is functionally equivalent to its counterpart, labs , in the standard header file <stdlib.h> .

Its format is also the same:

#include <stdlib.h> long int __LABS (long int x);

x

An integer.

This built-in function does, however, offer performance improvements because there is less call overhead associated with its use.

If you include <stdlib.h> , the built-in function is automatically used for all occurrences of labs . To disable the built-in function, use #undef labs .

Memory Barrier ( __MB)

The __MB function directs the compiler to generate a memory barrier instruction.

This function has the following format:

void __MB (void);

Memory Copy and Set Functions ( __MEMCPY, __MEMMOVE, __MEMSET)

The __MEMCPY, __MEMMOVE, and __MEMSET built-in functions are functionally equivalent to their counterparts in the standard header file <string.h> .

Their format is also the same:

#include <stdlib.h> void *__MEMCPY (void *s1, const void *s2, size_t size); void *__MEMMOVE (void *s1, const void *s2, size_t size); void *__MEMSET (void *s, int value, size_t size);

These built-in functions do, however, offer performance improvements because there is less call overhead associated with their use.

If you include <string.h> , the built-in functions are automatically used for all occurrences of memcpy , memmove , and memset . To disable the built-in functions, use #undef memcpy , #undef memmove , and #undef memset .

Read Process Cycle Counter ( __RPCC)

The __RPCC function reads the current process cycle counter.

This function has the following format:

int64 __RPCC (void);

Sine ( __SIN)

The __SIN built-in function is functionally equivalent to its counterpart in the standard header file <math.h> .

Its format is also the same:

#include <stdlib.h> double __SIN (double x);

x

A radian value.

This built-in function does, however, offer performance improvements because there is less call overhead associated with its use.

If you include <math.h> , the built-in function is used automatically for all occurrences of sin . To disable the built-in function, use #undef sin .

Test for Bit Clear then Clear Bit Interlocked ( __TESTBITCCI)

The __TESTBITCCI function performs the following operations in interlocked fashion:

• Returns the complement of the specified bit before being cleared
• Clears the bit

This function has the following format:

int __TESTBITCCI (void *address, int position, ...);

address

The base address of the field.

position

The position within the field of the bit that you want cleared.

...

An optional retry count of type int . If specified, the retry count indicates the number of times the operation is attempted. If the operation cannot be performed successfully in the specified number of retries, a value of 0 is returned.

Test for Bit Set then Set Bit Interlocked ( __TESTBITSSI)

The __TESTBITSSI function performs the following operations in interlocked fashion:

• Returns the value of the specified bit before being set
• Sets the bit

This function has the following format:

int __TESTBITSSI (void *address, int position, ...);

address

The base address of the field.

position

The position within the field of the bit that you want set.

...

An optional retry count of type int . If specified, the retry count indicates the number of times the operation is attempted. If the operation cannot be performed successfully in the specified number of retries, a value of 0 is returned.

Trap Barrier Instruction ( __TRAPB)

The __TRAPB function allows software to guarantee that, in a pipeline implementation, all previous arithmetic instructions will be completed without incurring any arithmetic traps before any instructions after the TRAPB instruction are issued.

This function has the following format:

void __TRAPB (void);

Unsigned Quadword Multiply High ( __UMULH)

The __UMULH function performs a quadword multiply high instruction.

This function has the following format:

uint64 __UMULH (uint64 operand1, uint64 operand2);

operand1

A 64-bit unsigned integer.

operand2

A 64-bit unsigned integer.

The two operands are multiplied as unsigned integers to produce a 128-bit result. The high order 64-bits are returned. Note that uint64 is a typedef for the HP Tru64 UNIX and Linux Alpha data type unsigned __int64 .

Other Builtins

int64 _popcnt(unsigned long);

Returns the number of "1" bits in the argument (0 to 64). For example, _popcnt(12) returns 2.

int64 _poppar(unsigned long);

Returns "1" if the number of "1" bits in the argument is odd; otherwise, returns "0". otherwise. For example: _poppar(12) returns 0.

int64 _leadz(unsigned long);

Returns the number of leading zeroes (starting at the most significant bit position) in the argument. For example, _leadz(1) returns 63. Note that _leadz(0) returns 64.

int64 _trailz(unsigned long);

Returns the number of trailing zeroes (counting after the least significant set bit to the least significant bit position) in the argument: For example, _trailz(2) returns 1. Note that _trailz(0) returns 64.