Current Fixes

Wrong answers were given at high opt levels with wide mode.

There was a problem running bbcomine unless software pipelining.

There was an occasional compiler assert in FE.

The compiler occsionally aborted with CHARACTER, POINTER, and DIMENSION initialized fields.

There was an Openmp private abort problem.

There was an occasional assert in the bridge.

At +O3, after inlining all uses of a module private routine, the compiler terminated with signal 11.

There was an occasional compiler abort at +O3.

Wrong answers were given at +O2 and above for MATMUL(TRANSPOSE(S),S).

There was a dataval name issue.

There was a rsqrt performance issue.

When an array index contained a variable with type qualifier, compiling at +O3 +DD64 would give an assert.

Two array declarations such as integer ar(100) and integer br(100) produced two unique array declarations, when only one was needed.

Nested subroutines produced erroneous addresses to break on when the compiler resumed generating code for the body of the containing routine.

Out-of-bounds access to an array stored in a common block yielded wrong answers at high optimization levels.

Wrong answers were occasionally received at +O3.

There was an issue with POINTER arrays.

Occasionally, declare_F90_F_STOP did not return.

SPECcpu2000 occasionally failed to collect the flow.data file.

There was a performance loss of implementation of a matrix multiplication.

Occasionally, there was a compiler internal error with OMP PARALLEL PRIVATE equivalence variables.

A module with a declared zero size would abort with +DD64.

The compiler occasionally asserted in FE.

There was a compiler abort with CHARACTER, POINTER, and DIMENSION initialized fields.

There was an Openmp private abort problem.

At +O3, after inlining all uses of a module private routine, the compiler terminated with signal 11.

There was an occasional compiler abort at +O3.

Wrong answers were given at +O2 and above for MATMUL(TRANSPOSE(S),S).

When using +O2, the compiler generated code which resulted in a memory core dump.

There was a dataval name issue.

There was a rsqrt performance issue.

The allocated memory size of common block differed by the way of declaration.

There was an occasional middle end assertion.

There was an issue with math inlining suppression for outlined parallel/workshare regions.

There was an HLO assertion failure at +O3.

When an array index contained a variable with type qualifier, compiling at +O3 +DD64 gave an assert.

Two array declarations such as integer ar(100) and integer br(100) produced two unique array declarations, when only one was needed.

Nested subroutines produced erroneous addresses to break on when the compiler resumed generating code for the body of the containing routine.