On Mon, Dec 12, 2011 at 11:49 PM, Jakub Jelinek <ja...@redhat.com> wrote: > On Mon, Dec 12, 2011 at 06:05:57PM -0800, Sriraman Tallam wrote: >> Do not vectorize loops on Core2 that need to use unaligned >> vector load/stores. >> * tree-vect-stmts.c (is_slow_vect_unaligned_load_store): New function. >> (vect_analyze_stmt): Check if the vectorizable load/store is slow. >> * target.def (TARGET_SLOW_UNALIGNED_VECTOR_MEMOP): New target hook. >> * doc/m.texi.in: Document new target hook: >> TARGET_SLOW_UNALIGNED_VECTOR_MEMOP >> * doc/m.texi: Regenerate. >> * config/i386/i386.c (ix86_slow_unaligned_vector_memop): New function. >> (TARGET_SLOW_UNALIGNED_VECTOR_MEMOP): New macro. > > IMHO it would be better if it didn't prevent vectorization of the loops > altogether, but lead to using aligned stores with an alignment check > before the vectorized loop if possible. > > Also, are unaligned loads equally expensive to unaligned stores?
Unaligned stores are always expensive, irrespective of whether the data turns out to be aligned or not at run-time. Unaligned loads are only as expensive when the data item is unaligned. For unaligned loads of aligned data, the movdqu is still slow but by ~2x rather than 6x. > > See http://gcc.gnu.org/PR49442 for further info, this really should be done > using some cost model rather than a boolean hook. > > Jakub
