| Message ID | CAAs8Hmz-_Hf8KFSLYO4OppjkVVLeAWU1qmnOF=C=dsXpXfEaPw@mail.gmail.com |
|---|---|
| State | New |
| Headers | show |
Can you collect some number on ggc_memory increase with this change in profile_gen build -- the value is recorded gcov_module_info object in coverage.c. thanks, David On Wed, Jul 24, 2013 at 3:14 PM, Sriraman Tallam <tmsriram@google.com> wrote: > The following test-case fails in LIPO mode during profile-use build: > > main.cc > ------------ > __attribute__((target("sse4.2"))) > unsigned int problem_aux (unsigned int A, unsigned int B); > > > int main (int argc, char *argv[]) > { > return problem_aux (0, 0); > } > > aux.cc > --------- > __attribute__((target("sse4.2"))) > unsigned int problem_aux (unsigned int A, unsigned int B) > { > return __builtin_ia32_crc32qi (A, B); > } > > $ g++ -O2 -fprofile-generate main.cc aux.cc > $ ./a.out > $ g++ -O2 -fprofile-use main.cc > error: '__builtin_ia32_crc32qi' was not declared in this scope > > This happens when parsing the auxiliary module. __builtin_ia32_crc32qi > is a target specific builtin that gets created after the target > attribute is seen in the declaration of problem_aux in the primary > module. This is too late for adding theisbuiltin to the list of > knowns, the buitlins get saved when cp_save_built_in_decl_pre_parsing > in cp/cp-objcp-common.c is called much earlier. Hence, this builtin > which is created during primary module parsing is not available when > parsing the auxiliary module. > > A simple fix is to expose all target builtins unconditionally in LIPO > mode. Patch: > > > Index: config/i386/i386.c > =================================================================== > --- config/i386/i386.c (revision 201046) > +++ config/i386/i386.c (working copy) > @@ -26957,7 +26957,8 @@ def_builtin (HOST_WIDE_INT mask, const char *name, > ix86_builtins_isa[(int) code].isa = mask; > > mask &= ~OPTION_MASK_ISA_64BIT; > - if (mask == 0 > + if (flag_dyn_ipa > + || mask == 0 > || (mask & ix86_isa_flags) != 0 > || (lang_hooks.builtin_function > == lang_hooks.builtin_function_ext_scope)) > > > Is this ok? > > Thanks > Sri
On Wed, Jul 24, 2013 at 3:21 PM, Xinliang David Li <davidxl@google.com> wrote: > Can you collect some number on ggc_memory increase with this change > in profile_gen build -- the value is recorded gcov_module_info object > in coverage.c. The ggc_memory increase if < 0.2% for large modules (initial usage around 100MB and greater). For some small modules, whose initial parser memory usage is around 2MB, I see a 80% increase. Thanks Sri > > thanks, > > David > > On Wed, Jul 24, 2013 at 3:14 PM, Sriraman Tallam <tmsriram@google.com> wrote: >> The following test-case fails in LIPO mode during profile-use build: >> >> main.ccthe >> ------------ >> __attribute__((target("sse4.2"))) >> unsigned int problem_aux (unsigned int A, unsigned int B); >> >> >> int main (int argc, char *argv[]) >> { >> return problem_aux (0, 0); >> } >> >> aux.cc >> --------- >> __attribute__((target("sse4.2"))) >> unsigned int problem_aux (unsigned int A, unsigned int B) >> { >> return __builtin_ia32_crc32qi (A, B); >> } >> >> $ g++ -O2 -fprofile-generate main.cc aux.cc >> $ ./a.out >> $ g++ -O2 -fprofile-use main.cc >> error: '__builtin_ia32_crc32qi' was not declared in this scope >> >> This happens when parsing the auxiliary module. __builtin_ia32_crc32qi >> is a target specific builtin that gets created after the target >> attribute is seen in the declaration of problem_aux in the primary >> module. This is too late for adding theisbuiltin to the list of >> knowns, the buitlins get saved when cp_save_built_in_decl_pre_parsing >> in cp/cp-objcp-common.c is called much earlier. Hence, this builtin >> which is created during primary module parsing is not available when >> parsing the auxiliary module. >> >> A simple fix is to expose all target builtins unconditionally in LIPO >> mode. Patch: >> >> >> Index: config/i386/i386.c >> =================================================================== >> --- config/i386/i386.c (revision 201046) >> +++ config/i386/i386.c (working copy) >> @@ -26957,7 +26957,8 @@ def_builtin (HOST_WIDE_INT mask, const char *name, >> ix86_builtins_isa[(int) code].isa = mask; >> >> mask &= ~OPTION_MASK_ISA_64BIT; >> - if (mask == 0 >> + if (flag_dyn_ipa >> + || mask == 0 >> || (mask & ix86_isa_flags) != 0 >> || (lang_hooks.builtin_function >> == lang_hooks.builtin_function_ext_scope)) >> >> >> Is this ok? >> >> Thanks >> Sri
the patch is ok. thanks, David On Thu, Jul 25, 2013 at 2:53 PM, Sriraman Tallam <tmsriram@google.com> wrote: > On Wed, Jul 24, 2013 at 3:21 PM, Xinliang David Li <davidxl@google.com> wrote: >> Can you collect some number on ggc_memory increase with this change >> in profile_gen build -- the value is recorded gcov_module_info object >> in coverage.c. > > The ggc_memory increase if < 0.2% for large modules (initial usage > around 100MB and greater). For some small modules, whose initial > parser memory usage is around 2MB, I see a 80% increase. > > Thanks > Sri > >> >> thanks, >> >> David >> >> On Wed, Jul 24, 2013 at 3:14 PM, Sriraman Tallam <tmsriram@google.com> wrote: >>> The following test-case fails in LIPO mode during profile-use build: >>> >>> main.ccthe >>> ------------ >>> __attribute__((target("sse4.2"))) >>> unsigned int problem_aux (unsigned int A, unsigned int B); >>> >>> >>> int main (int argc, char *argv[]) >>> { >>> return problem_aux (0, 0); >>> } >>> >>> aux.cc >>> --------- >>> __attribute__((target("sse4.2"))) >>> unsigned int problem_aux (unsigned int A, unsigned int B) >>> { >>> return __builtin_ia32_crc32qi (A, B); >>> } >>> >>> $ g++ -O2 -fprofile-generate main.cc aux.cc >>> $ ./a.out >>> $ g++ -O2 -fprofile-use main.cc >>> error: '__builtin_ia32_crc32qi' was not declared in this scope >>> >>> This happens when parsing the auxiliary module. __builtin_ia32_crc32qi >>> is a target specific builtin that gets created after the target >>> attribute is seen in the declaration of problem_aux in the primary >>> module. This is too late for adding theisbuiltin to the list of >>> knowns, the buitlins get saved when cp_save_built_in_decl_pre_parsing >>> in cp/cp-objcp-common.c is called much earlier. Hence, this builtin >>> which is created during primary module parsing is not available when >>> parsing the auxiliary module. >>> >>> A simple fix is to expose all target builtins unconditionally in LIPO >>> mode. Patch: >>> >>> >>> Index: config/i386/i386.c >>> =================================================================== >>> --- config/i386/i386.c (revision 201046) >>> +++ config/i386/i386.c (working copy) >>> @@ -26957,7 +26957,8 @@ def_builtin (HOST_WIDE_INT mask, const char *name, >>> ix86_builtins_isa[(int) code].isa = mask; >>> >>> mask &= ~OPTION_MASK_ISA_64BIT; >>> - if (mask == 0 >>> + if (flag_dyn_ipa >>> + || mask == 0 >>> || (mask & ix86_isa_flags) != 0 >>> || (lang_hooks.builtin_function >>> == lang_hooks.builtin_function_ext_scope)) >>> >>> >>> Is this ok? >>> >>> Thanks >>> Sri
Index: config/i386/i386.c =================================================================== --- config/i386/i386.c (revision 201046) +++ config/i386/i386.c (working copy) @@ -26957,7 +26957,8 @@ def_builtin (HOST_WIDE_INT mask, const char *name, ix86_builtins_isa[(int) code].isa = mask; mask &= ~OPTION_MASK_ISA_64BIT; - if (mask == 0 + if (flag_dyn_ipa + || mask == 0 || (mask & ix86_isa_flags) != 0 || (lang_hooks.builtin_function == lang_hooks.builtin_function_ext_scope))
The following test-case fails in LIPO mode during profile-use build: main.cc ------------ __attribute__((target("sse4.2"))) unsigned int problem_aux (unsigned int A, unsigned int B); int main (int argc, char *argv[]) { return problem_aux (0, 0); } aux.cc --------- __attribute__((target("sse4.2"))) unsigned int problem_aux (unsigned int A, unsigned int B) { return __builtin_ia32_crc32qi (A, B); } $ g++ -O2 -fprofile-generate main.cc aux.cc $ ./a.out $ g++ -O2 -fprofile-use main.cc error: '__builtin_ia32_crc32qi' was not declared in this scope This happens when parsing the auxiliary module. __builtin_ia32_crc32qi is a target specific builtin that gets created after the target attribute is seen in the declaration of problem_aux in the primary module. This is too late for adding theisbuiltin to the list of knowns, the buitlins get saved when cp_save_built_in_decl_pre_parsing in cp/cp-objcp-common.c is called much earlier. Hence, this builtin which is created during primary module parsing is not available when parsing the auxiliary module. A simple fix is to expose all target builtins unconditionally in LIPO mode. Patch: Is this ok? Thanks Sri