Cell Superscalar download

Solved compilation problems with GCC>=4.3

Worker runtime bugs
Master runtime
Compilation problems for C and Fortran
Installation directories problems fixed
Other bug fixes

Support for CellSim
Support for non aligned transfers
Compilation efficiency for diskless systems improved
Compiler support for libtool
File preprocessing in Fortran supported
Added some features to the compiler interface
Fortran compiler flexibility and eficiency improved
Worker functionality improvements
Improved worker traces

Several improvements in performance.
Added support for Itanium, AMD64 and the Altix achitecture.
Added experimental opaque type support (void * on the Cell).
Improved memory and internal object handling.
Use TMPDIR when generating intermediate traces if defined.
Modified aligned memory allocations to help valgrind.
Added an experimental and internal spin lock API.
Improved task stealing policy on SMP.
Run tasks in the main thread on SMP.
Css_malloc interface in the worker.
Improved stage-in code in the worker
Reduced fragmentation of the data buffer in the worker.

Fixed bug that causes some tasks to be executed twice.
Fixed renaming memory handling.
Fixed internal data being freed.
Fixed input scalar handling in SMP.
Fixed trace generation bugs.
Fixed race condition when renaming on a WaR dependency.
Fixed race condition where several processes compete to create a trace file.
Miscellaneous compiler fixes.
Fixed dma bug for >64K transfers.
Fixed various bugs in the worker library.

The master library has improved compilation speed.
The toolset can be compiled for use under a cross compiling environment.
Improved throughput by using better data alignment.
Bigger tracing buffers provide enhanced tracing performance.
Improved scheduling strategy with bundle prescheduling.
Task execution is pipelined in the SPUs even across bundles.
Improved latency by emmiting early SPU to PPU callbacks.
Improved SPU execution reliability thanks to stack overflow checks.
Better bandwith utilization by avoiding data transfers on accumulations.
General performance improvements in the PPU side.
The internal API has changed.

Improved native compiler error handling.
Fixed unsigned type handling in the compiler.
Fixed global variable handling.
Fixed constant parameter passing.
Fixed compiler configuration searching path.
Fixed "db16cyc" assembler instruction missing in Cell SDK 2.1.
Added compiler support for GCC statement expressions.