tree-optimization/107672 - avoid vector mode type_for_mode call
Checks
Commit Message
The following avoids using type_for_mode on vector modes which might
not work for all frontends. Instead we look for the inner mode
type and use build_vector_type_for_mode instead.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
PR tree-optimization/107672
* tree-vect-stmts.cc (supportable_widening_operation): Avoid
type_for_mode on vector modes.
---
gcc/tree-vect-stmts.cc | 12 +++++++++---
1 file changed, 9 insertions(+), 3 deletions(-)
Comments
Hi,
I am seeing an internal compiler error, related to this patch:
during GIMPLE pass: slp
options-save.cc: In function 'void cl_optimization_restore(gcc_options*, gcc_options*, cl_optimization*)':
options-save.cc:1292:1: internal compiler error: in supportable_widening_operation, at tree-vect-stmts.cc:12199
1292 | cl_optimization_restore (struct gcc_options *opts, struct gcc_options *opts_set,
| ^~~~~~~~~~~~~~~~~~~~~~~
/…/src/gcc/gcc/profile-count.cc: In member function 'int profile_count::to_cgraph_frequency(profile_count) const':
/…/src/gcc/gcc/profile-count.cc:308:1: note: parameter passing for argument of type 'profile_count' changed in GCC 9.1
308 | profile_count::to_cgraph_frequency (profile_count entry_bb_count) const
| ^~~~~~~~~~~~~
/…/src/gcc/gcc/profile-count.cc: In member function 'sreal profile_count::to_sreal_scale(profile_count, bool*) const':
/…/src/gcc/gcc/profile-count.cc:326:1: note: parameter passing for argument of type 'profile_count' changed in GCC 9.1
326 | profile_count::to_sreal_scale (profile_count in, bool *known) const
| ^~~~~~~~~~~~~
0x2195bdd supportable_widening_operation(vec_info*, tree_code, _stmt_vec_info*, tree_node*, tree_node*, tree_code*, tree_code*, int*, vec<tree_node*, va_heap, vl_ptr>*)
/…/src/gcc/gcc/tree-vect-stmts.cc:12199
0x2180493 vectorizable_conversion
/…/src/gcc/gcc/tree-vect-stmts.cc:5064
0x2192fdd vect_analyze_stmt(vec_info*, _stmt_vec_info*, bool*, _slp_tree*, _slp_instance*, vec<stmt_info_for_cost, va_heap, vl_ptr>*)
/…/src/gcc/gcc/tree-vect-stmts.cc:11256
/…/src/gcc/gcc/profile-count.cc: In member function 'profile_count profile_count::combine_with_ipa_count(profile_count)':
/…/src/gcc/gcc/profile-count.cc:398:1: note: parameter passing for argument of type 'profile_count' changed in GCC 9.1
398 | profile_count::combine_with_ipa_count (profile_count ipa)
| ^~~~~~~~~~~~~
0x14f95d1 vect_slp_analyze_node_operations_1
/…/src/gcc/gcc/tree-vect-slp.cc:5958
0x14f9c19 vect_slp_analyze_node_operations
/…/src/gcc/gcc/tree-vect-slp.cc:6147
0x14f9b4d vect_slp_analyze_node_operations
/…/src/gcc/gcc/tree-vect-slp.cc:6126
0x14fa439 vect_slp_analyze_operations(vec_info*)
/…/src/gcc/gcc/tree-vect-slp.cc:6387
0x14fd423 vect_slp_analyze_bb_1
/…/src/gcc/gcc/tree-vect-slp.cc:7372
0x14fd599 vect_slp_region
/…/src/gcc/gcc/tree-vect-slp.cc:7419
0x14fe0d1 vect_slp_bbs
/…/src/gcc/gcc/tree-vect-slp.cc:7610
0x14fe46f vect_slp_function(function*)
/…/src/gcc/gcc/tree-vect-slp.cc:7698
0x151a109 execute
/…/src/gcc/gcc/tree-vectorizer.cc:1532
Please submit a full bug report, with preprocessed source (by using -freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.
Makefile:1146: recipe for target 'options-save.o' failed
make[3]: *** [options-save.o] Error 1
That happens when building the arm-none-linux-gnueabihf toolchain natively with glibc bootstrap:
Build:arm-none-linux-gnueabihf
Host:arm-none-linux-gnueabihf
Target: arm-none-linux-gnueabihf
The compiler being used to build the toolchain is gcc 7.5.0.
Kind regards
Vasee
On Fri, 25 Nov 2022, Vaseeharan Vinayagamoorthy wrote:
> Hi,
>
> I am seeing an internal compiler error, related to this patch:
Can you please open a bugzilla for this and attach preprocessed
source so I can reproduce the ICE with a cc1 cross compiler?
Thanks,
Richard.
>
> during GIMPLE pass: slp
>
> options-save.cc: In function 'void cl_optimization_restore(gcc_options*,
> gcc_options*, cl_optimization*)':
>
> options-save.cc:1292:1: internal compiler error: in
> supportable_widening_operation, at tree-vect-stmts.cc:12199
>
>
>
> 1292 | cl_optimization_restore (struct gcc_options *opts, struct
> gcc_options *opts_set,
>
> | ^~~~~~~~~~~~~~~~~~~~~~~
>
> /…/src/gcc/gcc/profile-count.cc: In member function 'int
> profile_count::to_cgraph_frequency(profile_count) const':
>
> /…/src/gcc/gcc/profile-count.cc:308:1: note: parameter passing for argument
> of type 'profile_count' changed in GCC 9.1
>
> 308 | profile_count::to_cgraph_frequency (profile_count entry_bb_count)
> const
>
> | ^~~~~~~~~~~~~
>
> /…/src/gcc/gcc/profile-count.cc: In member function 'sreal
> profile_count::to_sreal_scale(profile_count, bool*) const':
>
> /…/src/gcc/gcc/profile-count.cc:326:1: note: parameter passing for argument
> of type 'profile_count' changed in GCC 9.1
>
> 326 | profile_count::to_sreal_scale (profile_count in, bool *known) const
>
> | ^~~~~~~~~~~~~
>
> 0x2195bdd supportable_widening_operation(vec_info*, tree_code,
> _stmt_vec_info*, tree_node*, tree_node*, tree_code*, tree_code*, int*,
> vec<tree_node*, va_heap, vl_ptr>*)
>
> /…/src/gcc/gcc/tree-vect-stmts.cc:12199
>
> 0x2180493 vectorizable_conversion
>
> /…/src/gcc/gcc/tree-vect-stmts.cc:5064
>
> 0x2192fdd vect_analyze_stmt(vec_info*, _stmt_vec_info*, bool*, _slp_tree*,
> _slp_instance*, vec<stmt_info_for_cost, va_heap, vl_ptr>*)
>
> /…/src/gcc/gcc/tree-vect-stmts.cc:11256
>
>
>
> /…/src/gcc/gcc/profile-count.cc: In member function 'profile_count
> profile_count::combine_with_ipa_count(profile_count)':
>
> /…/src/gcc/gcc/profile-count.cc:398:1: note: parameter passing for argument
> of type 'profile_count' changed in GCC 9.1
>
> 398 | profile_count::combine_with_ipa_count (profile_count ipa)
>
> | ^~~~~~~~~~~~~
>
> 0x14f95d1 vect_slp_analyze_node_operations_1
>
> /…/src/gcc/gcc/tree-vect-slp.cc:5958
>
> 0x14f9c19 vect_slp_analyze_node_operations
>
> /…/src/gcc/gcc/tree-vect-slp.cc:6147
>
> 0x14f9b4d vect_slp_analyze_node_operations
>
> /…/src/gcc/gcc/tree-vect-slp.cc:6126
>
> 0x14fa439 vect_slp_analyze_operations(vec_info*)
>
> /…/src/gcc/gcc/tree-vect-slp.cc:6387
>
> 0x14fd423 vect_slp_analyze_bb_1
>
> /…/src/gcc/gcc/tree-vect-slp.cc:7372
>
> 0x14fd599 vect_slp_region
>
> /…/src/gcc/gcc/tree-vect-slp.cc:7419
>
> 0x14fe0d1 vect_slp_bbs
>
> /…/src/gcc/gcc/tree-vect-slp.cc:7610
>
> 0x14fe46f vect_slp_function(function*)
>
> /…/src/gcc/gcc/tree-vect-slp.cc:7698
>
> 0x151a109 execute
>
> /…/src/gcc/gcc/tree-vectorizer.cc:1532
>
> Please submit a full bug report, with preprocessed source (by using
> -freport-bug).
>
> Please include the complete backtrace with any bug report.
>
> See <https://gcc.gnu.org/bugs/> for instructions.
>
> Makefile:1146: recipe for target 'options-save.o' failed
>
> make[3]: *** [options-save.o] Error 1
>
>
>
> That happens when building the arm-none-linux-gnueabihf toolchain natively
> with glibc bootstrap:
> Build:arm-none-linux-gnueabihf
> Host:arm-none-linux-gnueabihf
> Target: arm-none-linux-gnueabihf
>
> The compiler being used to build the toolchain is gcc 7.5.0.
>
>
> Kind regards
> Vasee
>
> ____________________________________________________________________________
> From: Gcc-patches <gcc-patches-bounces+vvinayag=arm.com@gcc.gnu.org> on
> behalf of Richard Biener via Gcc-patches <gcc-patches@gcc.gnu.org>
> Sent: 22 November 2022 08:48
> To: gcc-patches@gcc.gnu.org <gcc-patches@gcc.gnu.org>
> Subject: [PATCH] tree-optimization/107672 - avoid vector mode type_for_mode
> call
> The following avoids using type_for_mode on vector modes which might
> not work for all frontends. Instead we look for the inner mode
> type and use build_vector_type_for_mode instead.
>
> Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
>
> PR tree-optimization/107672
> * tree-vect-stmts.cc (supportable_widening_operation): Avoid
> type_for_mode on vector modes.
> ---
> gcc/tree-vect-stmts.cc | 12 +++++++++---
> 1 file changed, 9 insertions(+), 3 deletions(-)
>
> diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> index bc0ef136f19..b35b986889d 100644
> --- a/gcc/tree-vect-stmts.cc
> +++ b/gcc/tree-vect-stmts.cc
> @@ -12195,9 +12195,15 @@ supportable_widening_operation (vec_info *vinfo,
> intermediate_type
> = vect_halve_mask_nunits (prev_type, intermediate_mode);
> else
> - intermediate_type
> - = lang_hooks.types.type_for_mode (intermediate_mode,
> - TYPE_UNSIGNED (prev_type));
> + {
> + gcc_assert (VECTOR_MODE_P (intermediate_mode));
> + tree intermediate_element_type
> + = lang_hooks.types.type_for_mode (GET_MODE_INNER
> (intermediate_mode),
> + TYPE_UNSIGNED (prev_type));
> + intermediate_type
> + = build_vector_type_for_mode (intermediate_element_type,
> + intermediate_mode);
> + }
>
> if (VECTOR_BOOLEAN_TYPE_P (intermediate_type)
> && VECTOR_BOOLEAN_TYPE_P (prev_type)
> --
> 2.35.3
>
>
Hi Richard,
On 25/11/2022 21:08, Richard Biener via Gcc-patches wrote:
>
>
> On Fri, 25 Nov 2022, Vaseeharan Vinayagamoorthy wrote:
>
> > Hi,
> >
> > I am seeing an internal compiler error, related to this patch:
>
> Can you please open a bugzilla for this and attach preprocessed
> source so I can reproduce the ICE with a cc1 cross compiler?
I've raised a PR for this with a reduced testcase here:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107896
Thanks,
Alex
>
> Thanks,
> Richard.
>
> >
> > during GIMPLE pass: slp
> >
> > options-save.cc: In function 'void cl_optimization_restore(gcc_options*,
> > gcc_options*, cl_optimization*)':
> >
> > options-save.cc:1292:1: internal compiler error: in
> > supportable_widening_operation, at tree-vect-stmts.cc:12199
> >
> > �
> >
> > �1292 | cl_optimization_restore (struct gcc_options *opts, struct
> > gcc_options *opts_set,
> >
> > �| ^~~~~~~~~~~~~~~~~~~~~~~
> >
> > /�/src/gcc/gcc/profile-count.cc: In member function 'int
> > profile_count::to_cgraph_frequency(profile_count) const':
> >
> > /�/src/gcc/gcc/profile-count.cc:308:1: note: parameter passing for argument
> > of type 'profile_count' changed in GCC 9.1
> >
> > �308 | profile_count::to_cgraph_frequency (profile_count entry_bb_count)
> > const
> >
> > �| ^~~~~~~~~~~~~
> >
> > /�/src/gcc/gcc/profile-count.cc: In member function 'sreal
> > profile_count::to_sreal_scale(profile_count, bool*) const':
> >
> > /�/src/gcc/gcc/profile-count.cc:326:1: note: parameter passing for argument
> > of type 'profile_count' changed in GCC 9.1
> >
> > �326 | profile_count::to_sreal_scale (profile_count in, bool *known) const
> >
> > �| ^~~~~~~~~~~~~
> >
> > 0x2195bdd supportable_widening_operation(vec_info*, tree_code,
> > _stmt_vec_info*, tree_node*, tree_node*, tree_code*, tree_code*, int*,
> > vec<tree_node*, va_heap, vl_ptr>*)
> >
> > �/�/src/gcc/gcc/tree-vect-stmts.cc:12199
> >
> > 0x2180493 vectorizable_conversion
> >
> > �/�/src/gcc/gcc/tree-vect-stmts.cc:5064
> >
> > 0x2192fdd vect_analyze_stmt(vec_info*, _stmt_vec_info*, bool*, _slp_tree*,
> > _slp_instance*, vec<stmt_info_for_cost, va_heap, vl_ptr>*)
> >
> > �/�/src/gcc/gcc/tree-vect-stmts.cc:11256
> >
> > �
> >
> > /�/src/gcc/gcc/profile-count.cc: In member function 'profile_count
> > profile_count::combine_with_ipa_count(profile_count)':
> >
> > /�/src/gcc/gcc/profile-count.cc:398:1: note: parameter passing for argument
> > of type 'profile_count' changed in GCC 9.1
> >
> > �398 | profile_count::combine_with_ipa_count (profile_count ipa)
> >
> > �| ^~~~~~~~~~~~~
> >
> > 0x14f95d1 vect_slp_analyze_node_operations_1
> >
> > �/�/src/gcc/gcc/tree-vect-slp.cc:5958
> >
> > 0x14f9c19 vect_slp_analyze_node_operations
> >
> > �/�/src/gcc/gcc/tree-vect-slp.cc:6147
> >
> > 0x14f9b4d vect_slp_analyze_node_operations
> >
> > �/�/src/gcc/gcc/tree-vect-slp.cc:6126
> >
> > 0x14fa439 vect_slp_analyze_operations(vec_info*)
> >
> > �/�/src/gcc/gcc/tree-vect-slp.cc:6387
> >
> > 0x14fd423 vect_slp_analyze_bb_1
> >
> > �/�/src/gcc/gcc/tree-vect-slp.cc:7372
> >
> > 0x14fd599 vect_slp_region
> >
> > �/�/src/gcc/gcc/tree-vect-slp.cc:7419
> >
> > 0x14fe0d1 vect_slp_bbs
> >
> > �/�/src/gcc/gcc/tree-vect-slp.cc:7610
> >
> > 0x14fe46f vect_slp_function(function*)
> >
> > �/�/src/gcc/gcc/tree-vect-slp.cc:7698
> >
> > 0x151a109 execute
> >
> > �/�/src/gcc/gcc/tree-vectorizer.cc:1532
> >
> > Please submit a full bug report, with preprocessed source (by using
> > -freport-bug).
> >
> > Please include the complete backtrace with any bug report.
> >
> > See <https://gcc.gnu.org/bugs/> for instructions.
> >
> > Makefile:1146: recipe for target 'options-save.o' failed
> >
> > make[3]: *** [options-save.o] Error 1
> >
> >
> >
> > That happens when building the�arm-none-linux-gnueabihf toolchain natively
> > with glibc bootstrap:
> > Build:arm-none-linux-gnueabihf
> > Host:arm-none-linux-gnueabihf
> > Target: arm-none-linux-gnueabihf
> >
> > The compiler being used to build the toolchain is gcc 7.5.0.
> >
> >
> > Kind regards
> > Vasee
> >
> > ____________________________________________________________________________
> > From: Gcc-patches <gcc-patches-bounces+vvinayag=arm.com@gcc.gnu.org> on
> > behalf of Richard Biener via Gcc-patches <gcc-patches@gcc.gnu.org>
> > Sent: 22 November 2022 08:48
> > To: gcc-patches@gcc.gnu.org <gcc-patches@gcc.gnu.org>
> > Subject: [PATCH] tree-optimization/107672 - avoid vector mode type_for_mode
> > call �
> > The following avoids using type_for_mode on vector modes which might
> > not work for all frontends.� Instead we look for the inner mode
> > type and use build_vector_type_for_mode instead.
> >
> > Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
> >
> > ������� PR tree-optimization/107672
> > ������� * tree-vect-stmts.cc (supportable_widening_operation): Avoid
> > ������� type_for_mode on vector modes.
> > ---
> > �gcc/tree-vect-stmts.cc | 12 +++++++++---
> > �1 file changed, 9 insertions(+), 3 deletions(-)
> >
> > diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> > index bc0ef136f19..b35b986889d 100644
> > --- a/gcc/tree-vect-stmts.cc
> > +++ b/gcc/tree-vect-stmts.cc
> > @@ -12195,9 +12195,15 @@ supportable_widening_operation (vec_info *vinfo,
> > �������� intermediate_type
> > ���������� = vect_halve_mask_nunits (prev_type, intermediate_mode);
> > ������ else
> > -������ intermediate_type
> > -�������� = lang_hooks.types.type_for_mode (intermediate_mode,
> > -������������������������������������������ TYPE_UNSIGNED (prev_type));
> > +������ {
> > +�������� gcc_assert (VECTOR_MODE_P (intermediate_mode));
> > +�������� tree intermediate_element_type
> > +���������� = lang_hooks.types.type_for_mode (GET_MODE_INNER
> > (intermediate_mode),
> > +�������������������������������������������� TYPE_UNSIGNED (prev_type));
> > +�������� intermediate_type
> > +���������� = build_vector_type_for_mode (intermediate_element_type,
> > +���������������������������������������� intermediate_mode);
> > +������ }
> > �
> > ������ if (VECTOR_BOOLEAN_TYPE_P (intermediate_type)
> > ���������� && VECTOR_BOOLEAN_TYPE_P (prev_type)
> > --
> > 2.35.3
> >
> >
@@ -12195,9 +12195,15 @@ supportable_widening_operation (vec_info *vinfo,
intermediate_type
= vect_halve_mask_nunits (prev_type, intermediate_mode);
else
- intermediate_type
- = lang_hooks.types.type_for_mode (intermediate_mode,
- TYPE_UNSIGNED (prev_type));
+ {
+ gcc_assert (VECTOR_MODE_P (intermediate_mode));
+ tree intermediate_element_type
+ = lang_hooks.types.type_for_mode (GET_MODE_INNER (intermediate_mode),
+ TYPE_UNSIGNED (prev_type));
+ intermediate_type
+ = build_vector_type_for_mode (intermediate_element_type,
+ intermediate_mode);
+ }
if (VECTOR_BOOLEAN_TYPE_P (intermediate_type)
&& VECTOR_BOOLEAN_TYPE_P (prev_type)