Редагувати

Поділитися через


_control87, _controlfp, __control87_2

Gets and sets the floating-point control word. A more secure version of _controlfp is available; see _controlfp_s.

Syntax

unsigned int _control87(
   unsigned int new,
   unsigned int mask
);
unsigned int _controlfp(
   unsigned int new,
   unsigned int mask
);
int __control87_2(
   unsigned int new,
   unsigned int mask,
   unsigned int* x86_cw,
   unsigned int* sse2_cw
);

Parameters

new
New control-word bit values.

mask
Mask for new control-word bits to set.

x86_cw
Filled in with the control word for the x87 floating-point unit. Pass in 0 (NULL) to set only the SSE2 control word.

sse2_cw
Control word for the SSE floating-point unit. Pass in 0 (NULL) to set only the x87 control word.

Return value

For _control87 and _controlfp, the bits in the value returned indicate the floating-point control state. For a complete definition of the bits that are returned by _control87, see FLOAT.H.

For __control87_2, the return value is 1, which indicates success.

Remarks

The _control87 function gets and sets the floating-point control word. The floating-point control word enables the program to change the precision, rounding, and infinity modes, depending on the platform. You can also use _control87 to mask or unmask floating-point exceptions. If the value for mask is equal to 0, _control87 gets the floating-point control word. If mask is nonzero, a new value for the control word is set: For any bit that is on (that is, equal to 1) in mask, the corresponding bit in new is used to update the control word. In other words, fpcntrl = ((fpcntrl & ~mask) | (new & mask)) where fpcntrl is the floating-point control word.

Note

By default, the run-time libraries mask all floating-point exceptions.

_controlfp is a platform-independent, portable version of _control87 that's nearly identical to the _control87 function. If your code targets more than one platform, use _controlfp or _controlfp_s. The difference between _control87 and _controlfp is in how they treat DENORMAL values. For x86, x64, ARM, and ARM64 platforms, _control87 can set and clear the DENORMAL OPERAND exception mask. _controlfp doesn't modify the DENORMAL OPERAND exception mask. This example demonstrates the difference:

_control87( _EM_INVALID, _MCW_EM );
// DENORMAL is unmasked by this call
_controlfp( _EM_INVALID, _MCW_EM );
// DENORMAL exception mask remains unchanged

The possible values for the mask constant (mask) and new control values (new) are shown in the Control word masks and values table. Use the portable constants listed below (_MCW_EM, _EM_INVALID, and so forth) as arguments to these functions, rather than supplying the hexadecimal values explicitly.

Intel x86-derived platforms support the DENORMAL input and output values in hardware. The x86 behavior is to preserve DENORMAL values. The ARM and ARM64 platforms and the x64 platforms that have SSE2 support enable DENORMAL operands and results to be flushed, or forced to zero. The _controlfp and _control87 functions provide a mask to change this behavior. The following example demonstrates the use of this mask.

_controlfp(_DN_SAVE, _MCW_DN);
// Denormal values preserved on ARM platforms and on x64 processors with
// SSE2 support. NOP on x86 platforms.
_controlfp(_DN_FLUSH, _MCW_DN);
// Denormal values flushed to zero by hardware on ARM platforms
// and x64 processors with SSE2 support. Ignored on other x86 platforms.

On ARM and ARM64 platforms, the _control87 and _controlfp functions apply to the FPSCR register. Only the SSE2 control word that's stored in the MXCSR register is affected on x64 platforms. On x86 platforms, _control87 and _controlfp affect the control words for both the x87 and the SSE2, if present.

The function __control87_2 enables both the x87 and SSE2 floating-point units to be controlled together or separately. To affect both units, pass in the addresses of two integers to x86_cw and sse2_cw. If you only want to affect one unit, pass in an address for that parameter, but pass in 0 (NULL) for the other. If 0 is passed for one of these parameters, the function has no effect on that floating-point unit. It's useful when part of your code uses the x87 floating-point unit, and another part uses the SSE2 floating-point unit.

If you use __control87_2 to set different values for the floating-point control words, then _control87 or _controlfp might be unable to return a single control word to represent the state of both floating-point units. In such a case, these functions set the EM_AMBIGUOUS flag in the returned integer value to indicate an inconsistency between the two control words. The EM_AMBIGUOUS flag is a warning that the returned control word might not represent the state of both floating-point control words accurately.

On the ARM, ARM64, and x64 platforms, changing the infinity mode or the floating-point precision isn't supported. If the precision control mask is used on the x64 platform, the function raises an assertion, and the invalid parameter handler is invoked, as described in Parameter validation.

Note

__control87_2 is not supported on the ARM, ARM64, or x64 platforms. If you use __control87_2 and compile your program for the ARM, ARM64, or x64 platforms, the compiler generates an error.

These functions are ignored when you use /clr (Common Language Runtime Compilation) to compile. The common language runtime (CLR) only supports the default floating-point precision.

Control word masks and values

For the _MCW_EM mask, clearing the mask sets the exception, which allows the hardware exception; setting the mask hides the exception. If a _EM_UNDERFLOW or _EM_OVERFLOW occurs, no hardware exception is thrown until the next floating-point instruction is executed. To generate a hardware exception immediately after _EM_UNDERFLOW or _EM_OVERFLOW, call the FWAIT MASM instruction.

Mask Hex value Constant Hex value
_MCW_DN (Denormal control) 0x03000000 _DN_SAVE

_DN_FLUSH
0x00000000

0x01000000
_MCW_EM (Interrupt exception mask) 0x0008001F _EM_INVALID

_EM_DENORMAL

_EM_ZERODIVIDE

_EM_OVERFLOW

_EM_UNDERFLOW

_EM_INEXACT
0x00000010

0x00080000

0x00000008

0x00000004

0x00000002

0x00000001
_MCW_IC (Infinity control)

(Not supported on ARM or x64 platforms.)
0x00040000 _IC_AFFINE

_IC_PROJECTIVE
0x00040000

0x00000000
_MCW_RC (Rounding control) 0x00000300 _RC_CHOP

_RC_UP

_RC_DOWN

_RC_NEAR
0x00000300

0x00000200

0x00000100

0x00000000
_MCW_PC (Precision control)

(Not supported on ARM or x64 platforms.)
0x00030000 _PC_24 (24 bits)

_PC_53 (53 bits)

_PC_64 (64 bits)
0x00020000

0x00010000

0x00000000

Requirements

Routine Required header
_control87, _controlfp, _control87_2 <float.h>

For more compatibility information, see Compatibility.

Example

// crt_cntrl87.c
// processor: x86
// compile by using: cl /W4 /arch:IA32 crt_cntrl87.c
// This program uses __control87_2 to output the x87 control
// word, set the precision to 24 bits, and reset the status to
// the default.

#include <stdio.h>
#include <float.h>
#pragma fenv_access (on)

int main( void )
{
    double a = 0.1;
    unsigned int control_word_x87 = 0;
    int result;

    // Show original x87 control word and do calculation.
    result = __control87_2(0, 0, &control_word_x87, 0 );
    printf( "Original: 0x%.8x\n", control_word_x87 );
    printf( "%1.1f * %1.1f = %.15e\n", a, a, a * a );

    // Set precision to 24 bits and recalculate.
    result = __control87_2(_PC_24, MCW_PC, &control_word_x87, 0 );
    printf( "24-bit:   0x%.8x\n", control_word_x87 );
    printf( "%1.1f * %1.1f = %.15e\n", a, a, a * a );

    // Restore default precision-control bits and recalculate.
    result = __control87_2( _CW_DEFAULT, MCW_PC, &control_word_x87, 0 );
    printf( "Default:  0x%.8x\n", control_word_x87 );
    printf( "%1.1f * %1.1f = %.15e\n", a, a, a * a );
}
Original: 0x0009001f
0.1 * 0.1 = 1.000000000000000e-02
24-bit:   0x000a001f
0.1 * 0.1 = 9.999999776482582e-03
Default:  0x0009001f
0.1 * 0.1 = 1.000000000000000e-02

See also

Math and floating-point support
_clear87, _clearfp
_status87, _statusfp, _statusfp2
_controlfp_s