On Thu, May 26, 2011 at 12:52 AM, Ira Rosen <ira.ro...@linaro.org> wrote: > Hi, > > The vectorizer supports strided loads with gaps, e.g., when only a[4i] > and a[4i+2] are accessed, it generates a vector load a[4i:4i+3], i.e., > creating an access to a[4i+3], which doesn't exist in the scalar code. > This access maybe invalid as described in the PR. > > This patch creates an epilogue loop (with at least one iteration) for > such cases. > > Bootstrapped and tested on powerpc64-suse-linux. > Applied to trunk. I'll prepare patches for 4.5 and 4.6 next week. > > Ira > > > ChangeLog: > > PR tree-optimization/49038 > * tree-vect-loop-manip.c (vect_generate_tmps_on_preheader): > Ensure at least one epilogue iteration if required by data > accesses with gaps. > * tree-vectorizer.h (struct _loop_vec_info): Add new field > to mark loops that require peeling for gaps. > * tree-vect-loop.c (new_loop_vec_info): Initialize new field. > (vect_get_known_peeling_cost): Take peeling for gaps into > account. > (vect_transform_loop): Generate epilogue if required by data > access with gaps. > * tree-vect-data-refs.c (vect_analyze_group_access): Mark the > loop as requiring an epilogue if there are gaps in the end of > the strided group. > > testsuite/ChangeLog: > > PR tree-optimization/49038 > * gcc.dg/vect/vect-strided-u8-i8-gap4-unknown.c: New test.
This test fails at run-time at random on Linux/ia32: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49239 -- H.J.
