HP Fortran Version 3.3 Release Notes
Inline assembly support is desirable on Itanium based systems for the following reasons:
• To allow efficient source level access to machine resources, such as Application
registers, Control Registers, and Processor Status Register.
• To efficiently carry out low level operations such as synchronization, altering the
cache state and manipulating the register stack.
• To directly exploit powerful instructions that would otherwise be difficult for the
compiler to automatically generate from source level constructs.
Extended double-precision support for selected intrinsic functions
Real*10 (corresponding to float80 in C) is now supported for selected intrinsic
functions which include some of math intrinsic functions. It is an extended
double-precision format which helps to ensure accuracy during long chains of
computations involving double precision floating point numbers.
Selected intrinsic functions that have Real*10 support are:
• ACOS
• ACOSD
• ACOSH
• ATAN
• ATAND
• ATANH
• COS
• COSD
• COSH
• SIN
• SIND
• SINH
• SQRT
• RSQRT
Performance improvement in SUM intrinsic function
At +O2 or higher optimization levels the SUM intrinsic function is inlined with different
approach if:
• ARRAY argument is an array valued expression
• DIM argument is constant
• MASK is not present.
This enhancement improves the performance of the application containing SUMintrinsic
function call which satisfies the above three conditions.
8