Mathematical Acceleration Subsystem (MASS) libraries for AIX - previous versions

Download Description

Each successive MASS version includes the material from previous versions that has not been changed, except that: versions 3.0 and up do not include the library that runs only on POWER2 processors; versions 3.2 and up do not include the AIX 4 version of the POWER4 library; and versions 4.3 and up no longer include the AIX 4 libraries.

Version 1.0
- Introduces a library of mathematical intrinsic functions, libmass.a, that offers tuned alternatives for functions in the system math library supplied with AIX.

Version 2.1
- Adds two new libraries of vector function subroutines: libmassv.a for general use, and libmassvp2.a for use only on POWER2 machines. These vector subroutines compute result vectors as functions of their argument vectors.

Version 2.2
- Adds functions to the vector libraries.

Version 2.3
- Includes improved scalar sqrt and rsqrt functions.
- Adds more vector functions.
- Introduces a Fortran source file, libmassv.f, containing simple loops for the MASS vector functions that can be used on non-IBM systems.

Version 2.4
- Adds a tanh built-in function to the scalar library libmass.a.
- Slightly improves the sinh and x**y functions.
- Adds short-precision vector functions and vector atan2 to the general vector library libmassv.a.
- Revises the pow (C/C++) and x**y (Fortran) functions to accept negative x arguments with integer y arguments.

Version 2.5
- Makes the general vector library, libmassv.a, safe for multi-threaded applications where MASS functions might be called by two threads simultaneously.
- Further tunes some of these vector functions.
- Adds a new POWER3 vector library, libmassvp3.a (also thread-safe), with some routines tuned for the POWER3 architecture.
- Adds a dnint routine to the scalar library, libmass.a.

Version 2.6
- Makes the scalar library, libmass.a, safe for multi-threaded applications where MASS functions might be called by two threads simultaneously.
- Shows minor changes in performance of the scalar library.

Version 2.7
- Replaces twelve functions (forms of exp, log, sin, and cos) in the general vector library, libmassv.a, with versions rewritten for better short-precision performance and better performance for the POWER3 architecture.
- Includes the new functions in the architecture-specific vector libraries.

Version 3.0
- Augments the existing libmass.a, libmassv.a, and libmassvp3.a libraries with 64-bit versions of all functions, except vatan2 and vsatan2. The MASS libraries will now link to 32-bit or 64-bit applications.
- Adds a new POWER4 vector library, libmassvp4.a, that includes implementations of the vdiv, vsdiv, vrec, and vsrec functions that have been tuned for the POWER4 architecture.
- Removes the functions vdsign, vdfloat, and vidint from the libraries.

Version 3.2
- Adds POWER3- and POWER4-tuned versions of vacos and vasin to libmassvp3.a and libmassvp4.a, respectively.
- Adds POWER4-tuned versions of vlog, vexp, vrsqrt, vsqrt, and vtan, as well as their single-precision versions, vslog, vsexp, vsrsqrt, vssqrt, and vstan, to libmassv4.a.
- Fixes various bugs.
- Removes the AIX 4 version of libmassvp4.a (POWER4 machines do not run AIX 4).

Version 3.2.1
- Corrects some misspelled entry-point names.
- Updates the LICENSE and mass.html files.
- Adds some missing prototypes to massv.h.

Version 3.3
- Replaces vsqrt, vssqrt, vexp, vlog, vrec, and vdiv in libmassvp4.a with consistent versions.
- Improves the algorithms used by vrec and vdiv in libmassvp4.a.
- Adds new consistent POWER4-tuned vsin and vcos routines to libmassvp4.a. (There are some performance differences for short vectors.)

Version 3.4
- Adds the following new POWER4-tuned functions to libmassv.a, libmassvp3.a, and libmassvp4.a: vcosh, vscosh, vexpm1, vsexpm1, vlog10, vslog10, vlog1p, vslog1p, vpow, vspow, vsasin, vsacos, vsinh, vssinh, vtanh, vstanh.
- Improves the algorithms used by vsrec and vsdiv in libmassvp4.a to give better performance on GP POWER4 machines.
- Contains various bug fixes.
- Replaces the following routines in libmassvp4.a with consistent versions: vacos, vasin, vatan2, vrsqrt, vscos, vsdiv, vexp, vsrec, vssin. (There are some performance differences for short vectors.)

Version 4.1
- Adds the file mass.include to provide the Fortran interface block for the rsqrt scalar function.
- Adds the file mass.h to provide the C/C++ prototype for the dnint scalar function.
- Revises the vcosisin and vscosisin functions to allow use on AIX 4.x, where the _Complex data type is not supported.
- Adds the file libmassv.c to provide the vector source library code for C/C++.

Version 4.3
- Adds a new vector library libmassvp5.a that includes the following routines tuned for POWER5: vcbrt vrcbrt, vpopcnt4, vpopcnt8, vrqdrt, vqdrt, vrsqrt, vsqrt, vscbrt, vsrcbrt, vsrqdrt, vsqdrt, vsrsqrt, vssqrt. These new routines are also included in the other vector libraries.
- Improves the performance of vcbrt, vrcbrt, vqdrt, vscbrt, vsrcbrt, and vsqdrt in libmassv.a, libmassvp3.a, and libmassvp4.a.
- Adds new routines expm1 and log1p to libmass.a, and improves the performance of atan, cos, cosh, ** (Fortran), pow (C/C++), exp, sin, sinh, tan, and tanh in libmass.a.
- Adds the following routines, tuned for POWER4, to libmassvp4.a: atan, cosh, exp, expm1, log1p, pow, sin, cos, sinh, tan, tanh, vcbrt, vqdrt, vrcbrt,
- Contains various bug fixes.

Version 4.4
- Adds a new vector library libmassvp6.a that contains the full set of 60 MASS vector routines tuned for POWER6: vacos, vacosh, vasin, vasinh, vatan2, vatanh, vcbrt, vcos, vcosh, vcosisin, vdint, vdiv, vdnint, vexp, vexpm1, vlog, vlog10, vlog1p, vpopcnt4, vpopcnt8, vpow, vqdrt, vrcbrt, vrec, vrqdrt, vrsqrt, vsin, vsincos, vsinh, vsqrt, vtan, vtanh, vsacos, vsacosh, vsasin, vsasinh, vsatan2, vsatanh, vscbrt, vscos, vscosh, vscosisin, vsdiv, vsexp, vsexpm1, vslog, vslog10, vslog1p, vspow, vsqdrt, vsrcbrt, vsrec, vsrqdrt, vsrsqrt, vssin, vssincos, vssinh, vssqrt, vstan, vstanh.
- Adds 25 new single-precision scalar routines to libmass.a: acosf, acoshf, asinf, asinhf, atan2f, atanf, cbrtf, copysignf, atanhf, cosf, coshf, erfcf, erff, expf, expm1f, hypotf, lgammaf, log10f, log1pf, logf, powf, sinf, sinhf, tanf, tanhf.
- Adds 15 new double-precision scalar routines to libmass.a: acos, acosh, anint, asin, asinh, atanh, cbrt, copysign, cosisin, erf, erfc, hypot, lgamma, log10, sincos.
- Contains various bug fixes.

Version 6.0

- Adds a new vector library libmassvp7.a that contains the following routines tuned for POWER7 processors (functions new in this version are followed by *): vacos, vacosh, vasin, vasinh, vatan2, vatan*, vatanh, vcbrt, vcos, vcosisin, vcosh, vdint, vdiv, vdnint, verfc*, verf*, vexp2*, vexp2m1*, vexp, vexpm1, vhypot*, vlgamma*, vlog10, vlog1p, vlog2*, vlog21p*, vlog, vpow, vqdrt, vrcbrt, vrec, vrqdrt, vrsqrt, vsin, vsincos, vsinh, vsqrt, vtan, vtanh, vsacos, vsacosh, vsasin, vsasinh, vsatan2, vsatan*, vsatanh, vscbrt, vscos, vscosisin, vscosh, vsdiv, vserfc*, vserf*, vsexp2*, vsexp2m1*, vsexp, vsexpm1, vshypot*, vslog10, vslog1p, vslog2*, vslog21p*, vslog, vspow, vsqdrt, vsrcbrt, vsrec, vsrqdrt, vsrsqrt, vssin, vssincos, vssinh, vssqrt, vstan, vstanh, vpopcnt4, vpopcnt8.

- Adds a new SIMD library libmass_simdp7.a that contains the following routines tuned for POWER7 processors (all these functions are new in this version): acosd2, asind2, atan2d2, atand2, atanhd2, cbrtd2, cosd2, coshd2, divd2, erfcd2, erfd2, exp2d2, exp2m1d2, expd2, expm1d2, hypotd2, lgammad2, log10d2, log1pd2, log2d2, log21pd2, logd2, powd2, qdrtd2, rcbrtd2, recipd2, rqdrtd2, rsqrtd2, sind2,