These are the Release Notes for the AMD Vector Math Library.

Release 5 - June 23, 2006
The following functions are added in this release:  
  log2 vrd2_log2 vrd4_log2 vrda_log2
  fastlog2f vrs4_log2f vrs8_log2f vrsa_log2f
  vrsa_powf vrsa_powxf

Bug fixes:
1. Added a check to just return if N=0 in all array routines.
2. Added a .note.GNU-stack section to all linux routines to ensure
   a non-executable stack (security fix).
3. Added aliases in the Solaris version to support use of the scalar
   fastmath routines when linked with Fortran programs.
   These are added for exp, log, log10, pow, sin, cos, expf, logf,
   log10f, powf, sinf, and cosf.
   For example, the symbol __exp is aliased to fastexp.  This allows
   Fortran statements such as y=exp(x), where x and y are REAL*8 to
   use the fastexp routine.  Similarly, __expf is aliased to __fastexpf,
   allowing the statement using REAL*4 to use the fastexpf routine.

Release 4 - June 24, 2005
The following functions are added in this release:  
  fastsinf vrs4_sinf vrsa_sinf
  fastcosf vrs4_cosf vrsa_cosf
  fastsincosf vrs4_sincosf vrsa_sincosf
  fastlog10f vrs4_log10f vrs8_log10f vrsa_log10f
  vrs4_powxf

Bug fixes:
1. Fixed a problem in vrs4powf. rbx was not being saved as required by
the ABI.
2. Fixed a problem in fastsincos. r8 and rdx were not saved properly
across calls to remainder_piby2.  


Release 3 - April 14, 2005
The following functions are added to the release:  
  log10 vrd2_log10 vrd4_log10 vrda_log10
  fastpowf vrs4_powf

In addition, this release adds a acml_mv.h header file for use
with C and C++ programs. Three example files for C, C++, and fortran
usages are added.

Fortran interfaces were also added for the fast scalar and array
routines.  Fortran interfaces are not supplied for the vector 
functions since the _m128 types are not supported by current AMD64
Fortran compilers.  Examples of how to use the scalar and array
routines in a Fortran program are found in testfortranif.f.

All routines have been updated to be compatible with a shared object
library build.  Shared object build and testing was added to the AMD
internal validation suite.


Bug fixes:
1. All exp and expf routines would return the wrong value for very large
positive and very large negative input values (not including infinity).
Test cases were added to the AMD internal validation suite to test for
these cases.

2. Certain scalar routines were modified to prevent data in the upper
halves of xmm registers from causing exceptions.  These routines use
cvtpd2dq instructions, and NaNs in the upper half of the source register
could set the invalid operand exception.

Known issues:
none.


Release 2 - February 18, 2005
The following functions are added to the release:  
fastlogf vrs4logf vrs8logf vrsalogf


Release 1 - October 25, 2004
This is the first release of the AMD Vector Math Library source.
It includes the routines:
  fastlog vrd2log vrd4log vrdalog
  fastexp vrd2exp vrd4exp vrdaexp
  fastsin vrd2sin vrd4sin vrdasin
  fastcos vrd2cos vrd4cos vrdacos
  fastsincos vrd2sincos vrdasincos
  fastexpf vrs4expf vrs8expf vrsaexpf
  fastpow