| Message ID | OF9B23FCE9.37998EEB-ONC2257975.00452381-C2257975.004A77E7@il.ibm.com |
|---|---|
| State | New |
| Headers | show |
Thanks, I'll look into your remarks On Thu, Dec 29, 2011 at 5:33 PM, Ira Rosen <IRAR@il.ibm.com> wrote: > > > Igor Zamyatin <izamyatin@gmail.com> wrote on 29/12/2011 02:29:46 PM: > >> >> Because it includes AVX and AVX2 which deals with int and for AVX2 >> there are no problems with doubled diagnostics. > > And you can't just update vect_int because AVX does support it but with > 128-bit vectors, right? > So, your vect_float_no_int looks incorrect as well. > You need to describe the case when two vector sizes are analyzed, but the > first one always fails. Maybe vect_sizes_32B_16B_noint? Probably ugly, but > correct at least. > > I also suggest to simplify the checks and not to check the number of times > a pattern was detected , like this: > > Index: vect-widen-mult-half.c > =================================================================== > --- vect-widen-mult-half.c (revision 182703) > +++ vect-widen-mult-half.c (working copy) > @@ -43,7 +43,7 @@ int main (void) > } > > /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 > "vect" { target vect_widen_mult_hi_to_si } } } */ > -/* { dg-final { scan-tree-dump-times "vect_recog_widen_mult_pattern: > detected" 1 "vect" { target vect_widen_mult_hi_to_si_pattern } } } */ > +/* { dg-final { scan-tree-dump "vect_recog_widen_mult_pattern: detected" > "vect" { target vect_widen_mult_hi_to_si_pattern } } } */ > /* { dg-final { scan-tree-dump-times "pattern recognized" 1 > "vect" { target vect_widen_mult_hi_to_si_pattern } } } */ > /* { dg-final { cleanup-tree-dump "vect" } } */ > > > And split the tests with several loops into several files, like, for > example, vect-widen-mult-const-s16.c. > > The simplification is ok for slp-reduc-6.c, vect-119.c, vect-35-big-array.c > as well. > > For the rest of the tests, I don't understand why there are alignment > messages printed twice for two vector sizes. Why doesn't the vectorizer > fail during vect_determine_vectorization_factor? > > Ira > >> >> I understand that all this looks quite bulky but it's hard to create >> something which looks better without loosing generality >> >> >> On Thu, Dec 29, 2011 at 4:15 PM, Ira Rosen <IRAR@il.ibm.com> wrote: >> > >> > >> > Igor Zamyatin <izamyatin@gmail.com> wrote on 29/12/2011 02:04:45 PM: >> > >> >> When compiler configured with, say corei7-avx, it outputs twice more >> >> diagnostics on integer tests since AVX deals mostly with floats. I.e. >> >> compiler tries to vectorize on AVX vector size, than fails and then >> >> vectorizes on smaller vector size. This double work leads to double >> >> diagnostic output. >> > >> > OK, so you why not use vect_sizes_32B_16B? >> > >> > Ira >> > >> > >> >
Index: vect-widen-mult-half.c =================================================================== --- vect-widen-mult-half.c (revision 182703) +++ vect-widen-mult-half.c (working copy) @@ -43,7 +43,7 @@ int main (void) } /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target vect_widen_mult_hi_to_si } } } */ -/* { dg-final { scan-tree-dump-times "vect_recog_widen_mult_pattern: detected" 1 "vect" { target vect_widen_mult_hi_to_si_pattern } } } */ +/* { dg-final { scan-tree-dump "vect_recog_widen_mult_pattern: detected" "vect" { target vect_widen_mult_hi_to_si_pattern } } } */