This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] lno branch merge - vectorizer patch #3


Ira Rosen wrote:

Hi,

This patch adds vectorization support for additional forms of data
references along with relevant new test-cases.
We now support multidimensional arrays, pointer accesses with initial value
that is not restricted to an SSE_NAME (e.g: p=&a[16]-4B), and array
accesses with initial value that is not restricted to INTEGER_CST (e.g:
a[i+off]). Bootstrapped and passed the vectorizer testcases ans SPEC on
ppc-darwin.

I think you should also run the entire testsuite with the vectorizer hard-wired on. Have you done that?

+ #define VECT_SMODULO(x,y) ((x) % (y) < 0 ? ((x) % (y) + (y)) : (x) % (y))

This needs a comment.

! if (!evolution_function_is_constant_p (access_fn))
! {
! if (!vect_is_simple_iv_evolution (loop_containing_stmt (DR_STMT (dr))->num,
! access_fn, &init, &step, true))
! {
! if (vect_debug_details (NULL))


That should be:

 if (!e_f_i_c_p (access_fn)
      && !v_i_s_i_e (...)
      && vec_debug_details (NULL))

not three nested ifs. And vec_debug_details should probaby go at the outside since it will usually be false, avoiding computations implied by the other functions.

Patch OK with those changes.

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]