LLVM 3.2 Improves PowerPC Compiler Support
In addition to featuring an auto-vectorizer, Polly optimizations, and countless other improvements, the forthcoming release of LLVM 3.2 brings numerous improvements to its PowerPC back-end.
The PowerPC back-end target with LLVM 3.2 and accompanying Clang 3.2 C/C++ compiler feature many improvements for this compiler infrastructure that's due to be released in mid-December.
Among the PowerPC LLVM enhancements is better compliance with the 64-bit PowerPC ELF ABI, better inoperability with GCC, and a number of specific PowerPC64 feature improvements.
Some of the new LLVM PowerPC features include MCJIT support. PowerPC64 relocation support and TOC handling, various fixes, improvements to spilling and reloading for vector registers, C++ exception handling, fixes for big-endian code generation bugs, implemented -integrated-as support, and IBM long double support. With LLVM 3.2 is also 32/64-bit code generation improvements and instruction scheduling support for the FreeScale E500MC and E5500 cores.
For those interested in the LLVM/Clang x86 performance with the 3.2 SVN development code compared to GCC 4.8, there are recent compiler benchmarks. ARM-based compiler benchmarks are forthcoming.
The PowerPC back-end target with LLVM 3.2 and accompanying Clang 3.2 C/C++ compiler feature many improvements for this compiler infrastructure that's due to be released in mid-December.
Among the PowerPC LLVM enhancements is better compliance with the 64-bit PowerPC ELF ABI, better inoperability with GCC, and a number of specific PowerPC64 feature improvements.
Some of the new LLVM PowerPC features include MCJIT support. PowerPC64 relocation support and TOC handling, various fixes, improvements to spilling and reloading for vector registers, C++ exception handling, fixes for big-endian code generation bugs, implemented -integrated-as support, and IBM long double support. With LLVM 3.2 is also 32/64-bit code generation improvements and instruction scheduling support for the FreeScale E500MC and E5500 cores.
For those interested in the LLVM/Clang x86 performance with the 3.2 SVN development code compared to GCC 4.8, there are recent compiler benchmarks. ARM-based compiler benchmarks are forthcoming.
3 Comments