vrd4_log2: Four-valued double precision base-2 logarithm

• Prototype:
  • __m128d,__m128d __vrd4_log2(__m128d x1, __m128d x2);
    Note that this function uses a non-standard programming interface. The two __m128d inputs, which contain four double precision values, are passed by the AMD64 C ABI in registers xmm0, and xmm1. The corresponding results are returned in xmm0 and xmm1. The use of xmm1 to return a __m128d is non-standard, and this function can not be called directly from C. It can be called directly from assembly language. It is intended for internal use by vectorizing compilers, that may be able to take advantage of the non-standard calling interface.
  • Inputs:
    • __m128d x1 - the first double precision input value pair.
    • __m128d x2 - the second double precision input value pair.
  • Outputs:
    • The base-2 logarithm of x.
    • __m128d y1 - the first double precision base-2 logarithm result pair, returned in xmm0.
    • __m128d y2 - the second double precision base-2 logarithm result pair, returned in xmm1.
• Notes:
  • __vrd4_log2 computes the base-2 logarithm for each of four input arguments.
    
    This routine accepts four double precision input values passed as two __m128d values. The result is the double precision base-2 logarithm of the four values, returned as two __m128d values. This is a relaxed version of log2, suitable for use with fastmath compiler flags or applications not requiring full error handling. Denormal inputs may produce unpredictable results. Special case inputs produce C99 return values. The routine is accurate to better than 1 ulp over the valid input range.
• Special case return values:

  • Input	Output
    +/- 0	-infinity
    negative	QNaN
    QNaN	same QNaN
    SNaN	same NaN converted to QNaN
    +infinity	+infinity
    -infinity	QNaN

• Performance:
  • 235 cycles for most valid inputs (59 cycles per value), longer for input values very close to 1.0.