How Can I Efficiently Control Floating-Point Rounding Modes in C and x86 Assembly?

Efficiently Modifying Floating-Point Rounding in C and x86 Assembly

IEEE 754 floating-point numbers offer various rounding modes, such as nearest, zero, positive infinity, and negative infinity. Modifying the rounding mode enables precise control over the decimal representation of results.

C Solution

The C standard library provides a portable solution for altering rounding modes through the fesetround() function:

#include <fenv.h>
#pragma STDC FENV_ACCESS ON

int main() {
  // Save the original rounding mode
  int originalRounding = fegetround();

  // Set the desired rounding mode (e.g., to zero)
  fesetround(FE_TOWARDZERO);

  // Perform operations with the adjusted rounding mode

  // Restore the original rounding mode
  fesetround(originalRounding);

  return 0;
}</fenv.h>

x86 Assembly Solution

For platforms without C99 support, x86 assembly can be used to modify both the x87 unit and SSE rounding modes:

; x87 unit
fldcw [new rounding mode] ; e.g., FNINIT to nearest

; SSE
ldmxcsr [new rounding mode] ; e.g., MXCSR_RND_NEAREST

Microsoft Visual C

MSVC offers the non-standard _controlfp() function for this purpose:

#include <math.h>

int main() {
  // Save the original rounding mode
  unsigned int originalRounding = _controlfp(0, 0);

  // Set the desired rounding mode (e.g., to zero)
  _controlfp(_RC_CHOP, _MCW_RC);

  // Perform operations with the adjusted rounding mode

  // Restore the original rounding mode
  _controlfp(originalRounding, _MCW_RC);

  return 0;
}</math.h>

Rounding Mode Decoder Ring

Rounding Mode	C Name	MSVC Name
Nearest	FE_TONEAREST	_RC_NEAR
Zero	FE_TOWARDZERO	_RC_CHOP
Infinity	FE_UPWARD	_RC_UP
-Infinity	FE_DOWNWARD	_RC_DOWN

The above is the detailed content of How Can I Efficiently Control Floating-Point Rounding Modes in C and x86 Assembly?. For more information, please follow other related articles on the PHP Chinese website!