Re: [PATCH] Fix *vector_shift_pattern (PR tree-optimization/70354)

2016-03-23 Thread Richard Biener
On Tue, 22 Mar 2016, Jakub Jelinek wrote: > Hi! > > As the testcase shows, the C/C++ FEs narrow the shift counters from whatever > type they had originally to unsigned int (previously signed int). > Then the vect-patterns code, to be able to use vector by vector shifts > attempts to narrow or wid

[PATCH] Fix *vector_shift_pattern (PR tree-optimization/70354)

2016-03-22 Thread Jakub Jelinek
Hi! As the testcase shows, the C/C++ FEs narrow the shift counters from whatever type they had originally to unsigned int (previously signed int). Then the vect-patterns code, to be able to use vector by vector shifts attempts to narrow or widen them again to the right type. If there is already a