Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Microsoft Specific
Emits the Streaming SIMD Extensions 4 (SSE4) instruction roundps. This instruction rounds a 32 bit value by using the specified rounding control.
__m128 _mm_round_ps(
__m128 a,
const int cntrl
);
Parameters
[in] a
A 128-bit parameter that contains four 32-bit floating point values.[in] cntrl
A constant that specifies control fields for the rounding operation.
Return value
A 128-bit parameter. The result can be expressed with the following equations:.
r0 := RND(a0)
r1 := RND(a1)
r2 := RND(a2)
r3 := RND(a3)
Requirements
Intrinsic |
Architecture |
|---|---|
_mm_round_ps |
x86, x64 |
Header file <smmintrin.h>
Remarks
r0-r3, a0-a3, and b0-b3 are the sequentially ordered 32-bit components of return value r and parameters a and b, respectively. r0, a0, and b0 are the least significant 32 bits.
The rounding function uses the cntrl parameter to determine how to compute a new value. The following table indicates what rounding mode will be used.
Rounding mode |
Value |
Description |
|---|---|---|
_MM_FROUND_TO_NEAREST_INT |
0x0 |
Round to nearest (even). |
_MM_FROUND_TO_NEG_INF |
0x1 |
Round down (toward -∞). |
_MM_FROUND_TO_POS_INF |
0x2 |
Round up (toward +∞). |
_MM_FROUND_TO_ZERO |
0x3 |
Round toward zero (truncate). |
_MM_FROUND_CUR_DIRECTION |
0x4 |
Use current MXCSR setting. |
This table shows how cntrl determines whether an exception should be signaled when a SNaN is detected.
Precision exception handling |
Value |
Description |
|---|---|---|
_MM_FROUND_RAISE_EXC |
0x0 |
Signal precision exception on SNaN. |
_MM_FROUND_NO_EXC |
0x8 |
Do not signal precision exception on SNaN. |
The following macros are also available to combine the above two fields:
Rounding mode and precision exception handling |
Value |
|---|---|
_MM_FROUND_NINT |
MM_FROUND_TO_NEAREST_INT | _MM_FROUND_RAISE_EXC |
_MM_FROUND_FLOOR |
_MM_FROUND_TO_NEG_INF | _MM_FROUND_RAISE_EXC |
_MM_FROUND_CEIL |
_MM_FROUND_TO_POS_INF | _MM_FROUND_RAISE_EXC |
_MM_FROUND_TRUNC |
_MM_FROUND_TO_ZERO | _MM_FROUND_RAISE_EXC |
_MM_FROUND_RINT |
_MM_FROUND_CUR_DIRECTION | _MM_FROUND_RAISE_EXC |
_MM_FROUND_NEARBYINT |
_MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC |
Before you use this intrinsic, software must ensure that the underlying processor supports the instruction.
Example
#include <stdio.h>
#include <smmintrin.h>
int main ()
{
__m128 a;
const int cntrl = _MM_FROUND_FLOOR;
a.m128_f32[0] = 9.9375;
a.m128_f32[1] = 5964.125;
a.m128_f32[2] = -237.875;
a.m128_f32[3] = -0.125;
__m128 res = _mm_round_ps(a, cntrl);
printf_s("Original a: %f\t%f\t%f\t%f\n",
a.m128_f32[0], a.m128_f32[1], a.m128_f32[2], a.m128_f32[3]);
printf_s("Result res: %f\t%f\t%f\t%f\n",
res.m128_f32[0], res.m128_f32[1], res.m128_f32[2], res.m128_f32[3]);
return 0;
}
Original a: 9.937500 5964.125000 -237.875000 -0.125000 Result res: 9.000000 5964.000000 -238.000000 -1.000000