On Monday, October 11, 2021, Jakub Jelinek via Gcc-patches < gcc-patches@gcc.gnu.org> wrote:
> On Mon, Oct 11, 2021 at 11:28:55AM +0200, Jakub Jelinek via Gcc-patches > wrote: > > I don't think this is the right thing to do. > > This just means that at some point between 2013 when -fsimd-cost-model > has > > been introduced and now -fsimd-cost-model= option at least partially > stopped > > working properly. > > As documented, -fsimd-cost-model= overrides the -fvect-cost-model= > setting > > for OpenMP simd loops (loop->force_vectorize is true) if specified > differently > > from default. > > In tree-vectorizer.h we have: > > static inline bool > > unlimited_cost_model (loop_p loop) > > { > > if (loop != NULL && loop->force_vectorize > > && flag_simd_cost_model != VECT_COST_MODEL_DEFAULT) > > return flag_simd_cost_model == VECT_COST_MODEL_UNLIMITED; > > return (flag_vect_cost_model == VECT_COST_MODEL_UNLIMITED); > > } > > and use it in various places, but we also just use flag_vect_cost_model > > in lots of places (and in one spot use flag_simd_cost_model, not sure if > > we are sure it is a force_vectorize loop or what). > > > > So, IMHO we should change the above inline function to > > loop_cost_model and let it return the cost model and then just > > reimplement unlimited_cost_model as > > return loop_cost_model (loop) == VECT_COST_MODEL_UNLIMITED; > > and then adjust the direct uses of the flag and revert these changes. > > Here is a patch that implements it. > Ok for trunk if it passes bootstrap/regtest? > > 2021-10-11 Jakub Jelinek <ja...@redhat.com> > > gcc/ > * tree-vectorizer.h (loop_cost_model): New function. > (unlimited_cost_model): Use it. > * tree-vect-loop.c (vect_analyze_loop_costing): Use loop_cost_model > call instead of flag_vect_cost_model. > * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): > Likewise. > (vect_prune_runtime_alias_test_list): Likewise. Also use it > instead > of flag_simd_cost_model. > libgomp/ > * testsuite/libgomp.c/scan-11.c: Remove option > -fvect-cost-model=cheap. > * testsuite/libgomp.c/scan-12.c: Likewise. > * testsuite/libgomp.c/scan-13.c: Likewise. > * testsuite/libgomp.c/scan-14.c: Likewise. > * testsuite/libgomp.c/scan-15.c: Likewise. > * testsuite/libgomp.c/scan-16.c: Likewise. > * testsuite/libgomp.c/scan-17.c: Likewise. > * testsuite/libgomp.c/scan-18.c: Likewise. > * testsuite/libgomp.c/scan-19.c: Likewise. > * testsuite/libgomp.c/scan-20.c: Likewise. > * testsuite/libgomp.c/scan-21.c: Likewise. > * testsuite/libgomp.c/scan-22.c: Likewise. > * testsuite/libgomp.c++/scan-9.C: Likewise. > * testsuite/libgomp.c++/scan-10.C: Likewise. > * testsuite/libgomp.c++/scan-11.C: Likewise. > * testsuite/libgomp.c++/scan-12.C: Likewise. > * testsuite/libgomp.c++/scan-13.C: Likewise. > * testsuite/libgomp.c++/scan-14.C: Likewise. > * testsuite/libgomp.c++/scan-15.C: Likewise. > * testsuite/libgomp.c++/scan-16.C: Likewise. Also for gcc.dg/gomp/simd-2.c, gcc.dg/gomp/simd-3.c --- gcc/tree-vectorizer.h.jj 2021-09-27 10:47:15.839084866 +0200 > +++ gcc/tree-vectorizer.h 2021-10-11 13:46:55.169767481 +0200 > @@ -1701,14 +1701,22 @@ get_dr_vinfo_offset (vec_info *vinfo, > } > > > +/* Return the vect cost model for LOOP. */ > +static inline enum vect_cost_model > +loop_cost_model (loop_p loop) > +{ > + if (loop != NULL > + && loop->force_vectorize > + && flag_simd_cost_model != VECT_COST_MODEL_DEFAULT) > + return flag_simd_cost_model; > + return flag_vect_cost_model; > +} > + > /* Return true if the vect cost model is unlimited. */ > static inline bool > unlimited_cost_model (loop_p loop) > { > - if (loop != NULL && loop->force_vectorize > - && flag_simd_cost_model != VECT_COST_MODEL_DEFAULT) > - return flag_simd_cost_model == VECT_COST_MODEL_UNLIMITED; > - return (flag_vect_cost_model == VECT_COST_MODEL_UNLIMITED); > + return loop_cost_model (loop) == VECT_COST_MODEL_UNLIMITED; > } > > /* Return true if the loop described by LOOP_VINFO is fully-masked and > --- gcc/tree-vect-loop.c.jj 2021-09-22 09:25:15.199030463 +0200 > +++ gcc/tree-vect-loop.c 2021-10-11 13:48:33.183366790 +0200 > @@ -1850,7 +1850,7 @@ vect_analyze_loop_costing (loop_vec_info > > /* If using the "very cheap" model. reject cases in which we'd keep > a copy of the scalar code (even if we might be able to vectorize > it). */ > - if (flag_vect_cost_model == VECT_COST_MODEL_VERY_CHEAP > + if (loop_cost_model (loop) == VECT_COST_MODEL_VERY_CHEAP > && (LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) > || LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo) > || LOOP_VINFO_PEELING_FOR_NITER (loop_vinfo))) > @@ -1922,7 +1922,7 @@ vect_analyze_loop_costing (loop_vec_info > /* If the vector loop needs multiple iterations to be beneficial then > things are probably too close to call, and the conservative thing > would be to stick with the scalar code. */ > - if (flag_vect_cost_model == VECT_COST_MODEL_VERY_CHEAP > + if (loop_cost_model (loop) == VECT_COST_MODEL_VERY_CHEAP > && min_profitable_estimate > (int) vect_vf_for_cost (loop_vinfo)) > { > if (dump_enabled_p ()) > --- gcc/tree-vect-data-refs.c.jj 2021-09-30 17:12:15.238586878 +0200 > +++ gcc/tree-vect-data-refs.c 2021-10-11 13:51:48.491576058 +0200 > @@ -2236,7 +2236,7 @@ vect_enhance_data_refs_alignment (loop_v > { > unsigned max_allowed_peel > = param_vect_max_peeling_for_alignment; > - if (flag_vect_cost_model <= VECT_COST_MODEL_CHEAP) > + if (loop_cost_model (loop) <= VECT_COST_MODEL_CHEAP) > max_allowed_peel = 0; > if (max_allowed_peel != (unsigned)-1) > { > @@ -2334,7 +2334,7 @@ vect_enhance_data_refs_alignment (loop_v > do_versioning > = (optimize_loop_nest_for_speed_p (loop) > && !loop->inner /* FORNOW */ > - && flag_vect_cost_model > VECT_COST_MODEL_CHEAP); > + && loop_cost_model (loop) > VECT_COST_MODEL_CHEAP); > > if (do_versioning) > { > @@ -3751,7 +3751,9 @@ vect_prune_runtime_alias_test_list (loop > unsigned int count = (comp_alias_ddrs.length () > + check_unequal_addrs.length ()); > > - if (count && flag_vect_cost_model == VECT_COST_MODEL_VERY_CHEAP) > + if (count > + && (loop_cost_model (LOOP_VINFO_LOOP (loop_vinfo)) > + == VECT_COST_MODEL_VERY_CHEAP)) > return opt_result::failure_at > (vect_location, "would need a runtime alias check\n"); > > @@ -3760,7 +3762,7 @@ vect_prune_runtime_alias_test_list (loop > "improved number of alias checks from %d to %d\n", > may_alias_ddrs.length (), count); > unsigned limit = param_vect_max_version_for_alias_checks; > - if (flag_simd_cost_model == VECT_COST_MODEL_CHEAP) > + if (loop_cost_model (LOOP_VINFO_LOOP (loop_vinfo)) == > VECT_COST_MODEL_CHEAP) > limit = param_vect_max_version_for_alias_checks * 6 / 10; > if (count > limit) > return opt_result::failure_at > --- libgomp/testsuite/libgomp.c/scan-11.c.jj 2021-10-10 > 12:12:28.366333440 +0200 > +++ libgomp/testsuite/libgomp.c/scan-11.c 2021-10-11 > 13:53:36.090038707 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fvect-cost-model=cheap -fopenmp > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-12.c.jj 2021-10-10 > 12:12:28.366333440 +0200 > +++ libgomp/testsuite/libgomp.c/scan-12.c 2021-10-11 > 13:53:40.186980174 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-13.c.jj 2021-10-10 > 12:12:28.366333440 +0200 > +++ libgomp/testsuite/libgomp.c/scan-13.c 2021-10-11 > 13:53:43.538932282 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-14.c.jj 2021-10-10 > 12:12:28.366333440 +0200 > +++ libgomp/testsuite/libgomp.c/scan-14.c 2021-10-11 > 13:53:46.637888006 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-15.c.jj 2021-10-10 > 12:12:28.366333440 +0200 > +++ libgomp/testsuite/libgomp.c/scan-15.c 2021-10-11 > 13:53:49.891841513 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-16.c.jj 2021-10-10 > 12:12:28.367333426 +0200 > +++ libgomp/testsuite/libgomp.c/scan-16.c 2021-10-11 > 13:53:52.908798405 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-17.c.jj 2021-10-10 > 12:12:28.367333426 +0200 > +++ libgomp/testsuite/libgomp.c/scan-17.c 2021-10-11 > 13:53:55.940755086 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-18.c.jj 2021-10-10 > 12:12:28.367333426 +0200 > +++ libgomp/testsuite/libgomp.c/scan-18.c 2021-10-11 > 13:53:58.739715092 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-19.c.jj 2021-10-10 > 12:12:28.367333426 +0200 > +++ libgomp/testsuite/libgomp.c/scan-19.c 2021-10-11 > 13:54:01.640673646 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-20.c.jj 2021-10-10 > 12:12:28.367333426 +0200 > +++ libgomp/testsuite/libgomp.c/scan-20.c 2021-10-11 > 13:54:05.757614824 +0200 > @@ -1,5 +1,5 @@ > /* { dg-require-effective-target size32plus } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } */ > /* { dg-additional-options "-msse2" { target sse2_runtime } } */ > /* { dg-additional-options "-mavx" { target avx_runtime } } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } */ > --- libgomp/testsuite/libgomp.c/scan-21.c.jj 2021-10-10 > 12:12:28.367333426 +0200 > +++ libgomp/testsuite/libgomp.c/scan-21.c 2021-10-11 > 13:54:09.852556318 +0200 > @@ -1,6 +1,6 @@ > /* { dg-require-effective-target size32plus } */ > /* { dg-require-effective-target avx_runtime } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details -msse2 -mno-sse3" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details -msse2 > -mno-sse3" } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" } > } */ > > #include "scan-13.c" > --- libgomp/testsuite/libgomp.c/scan-22.c.jj 2021-10-10 > 12:12:28.367333426 +0200 > +++ libgomp/testsuite/libgomp.c/scan-22.c 2021-10-11 > 13:54:14.516489679 +0200 > @@ -1,6 +1,6 @@ > /* { dg-require-effective-target size32plus } */ > /* { dg-require-effective-target avx_runtime } */ > -/* { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details -msse2 -mno-sse3" } */ > +/* { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details -msse2 > -mno-sse3" } */ > /* { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" } > } */ > > #include "scan-17.c" > --- libgomp/testsuite/libgomp.c++/scan-9.C.jj 2021-10-10 > 12:12:28.000000000 +0200 > +++ libgomp/testsuite/libgomp.c++/scan-9.C 2021-10-11 > 13:54:47.256021904 +0200 > @@ -1,5 +1,5 @@ > // { dg-require-effective-target size32plus } > -// { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } > +// { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } > // { dg-additional-options "-msse2" { target sse2_runtime } } > // { dg-additional-options "-mavx" { target avx_runtime } } > // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } > --- libgomp/testsuite/libgomp.c++/scan-10.C.jj 2021-10-10 > 12:12:28.000000000 +0200 > +++ libgomp/testsuite/libgomp.c++/scan-10.C 2021-10-11 > 13:54:50.123980925 +0200 > @@ -1,5 +1,5 @@ > // { dg-require-effective-target size32plus } > -// { dg-additional-options "-O2 -fvect-cost-model=cheap -fopenmp > -fdump-tree-vect-details" } > +// { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } > // { dg-additional-options "-msse2" { target sse2_runtime } } > // { dg-additional-options "-mavx" { target avx_runtime } } > // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } > --- libgomp/testsuite/libgomp.c++/scan-11.C.jj 2021-10-10 > 12:12:28.000000000 +0200 > +++ libgomp/testsuite/libgomp.c++/scan-11.C 2021-10-11 > 13:54:54.256921875 +0200 > @@ -1,5 +1,5 @@ > // { dg-require-effective-target size32plus } > -// { dg-additional-options "-O2 -fvect-cost-model=cheap -fopenmp > -fdump-tree-vect-details" } > +// { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } > // { dg-additional-options "-msse2" { target sse2_runtime } } > // { dg-additional-options "-mavx" { target avx_runtime } } > // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } > --- libgomp/testsuite/libgomp.c++/scan-12.C.jj 2021-10-10 > 12:12:28.000000000 +0200 > +++ libgomp/testsuite/libgomp.c++/scan-12.C 2021-10-11 > 13:54:57.077881570 +0200 > @@ -1,5 +1,5 @@ > // { dg-require-effective-target size32plus } > -// { dg-additional-options "-O2 -fvect-cost-model=cheap -fopenmp > -fdump-tree-vect-details" } > +// { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } > // { dg-additional-options "-msse2" { target sse2_runtime } } > // { dg-additional-options "-mavx" { target avx_runtime } } > // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } > --- libgomp/testsuite/libgomp.c++/scan-13.C.jj 2021-10-10 > 12:12:28.000000000 +0200 > +++ libgomp/testsuite/libgomp.c++/scan-13.C 2021-10-11 > 13:55:00.164837465 +0200 > @@ -1,5 +1,5 @@ > // { dg-require-effective-target size32plus } > -// { dg-additional-options "-O2 -fvect-cost-model=cheap -fopenmp > -fdump-tree-vect-details" } > +// { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } > // { dg-additional-options "-msse2" { target sse2_runtime } } > // { dg-additional-options "-mavx" { target avx_runtime } } > // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } > --- libgomp/testsuite/libgomp.c++/scan-14.C.jj 2021-10-10 > 12:12:28.000000000 +0200 > +++ libgomp/testsuite/libgomp.c++/scan-14.C 2021-10-11 > 13:55:03.960783227 +0200 > @@ -1,5 +1,5 @@ > // { dg-require-effective-target size32plus } > -// { dg-additional-options "-O2 -fvect-cost-model=cheap -fopenmp > -fdump-tree-vect-details" } > +// { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } > // { dg-additional-options "-msse2" { target sse2_runtime } } > // { dg-additional-options "-mavx" { target avx_runtime } } > // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } > --- libgomp/testsuite/libgomp.c++/scan-15.C.jj 2021-10-10 > 12:12:28.000000000 +0200 > +++ libgomp/testsuite/libgomp.c++/scan-15.C 2021-10-11 > 13:55:06.864741739 +0200 > @@ -1,5 +1,5 @@ > // { dg-require-effective-target size32plus } > -// { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } > +// { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } > // { dg-additional-options "-msse2" { target sse2_runtime } } > // { dg-additional-options "-mavx" { target avx_runtime } } > // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } > --- libgomp/testsuite/libgomp.c++/scan-16.C.jj 2021-10-10 > 12:12:28.000000000 +0200 > +++ libgomp/testsuite/libgomp.c++/scan-16.C 2021-10-11 > 13:55:09.861698918 +0200 > @@ -1,5 +1,5 @@ > // { dg-require-effective-target size32plus } > -// { dg-additional-options "-O2 -fopenmp -fvect-cost-model=cheap > -fdump-tree-vect-details" } > +// { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" } > // { dg-additional-options "-msse2" { target sse2_runtime } } > // { dg-additional-options "-mavx" { target avx_runtime } } > // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { > target sse2_runtime } } } > > > Jakub > > -- BR, Hongtao