Skip to contentSkip to conversation

glibc: Help compile with -Os

PATCH In review Khem Raj 4 participants 8 messages Feb 24, 2015 – Mar 6, 2015

lore
Share & tools
lore

Khem Raj

author
#0 lore raw

PATCH 1/1 glibc: Help compile with -Os

When we modify to use -Os
-Werror doesnt go well with it, glibc needs to be cleaned up for that but until then lets disable -Werror when using -Os

Change-Id I5495255fce67953f15c07e423e3e0eef41d4ce5e
---
 meta/recipes-core/glibc/glibc.inc                  |  3 ++
 .../glibc/glibc/eglibc-use-option-groups.patch     | 47 ++++++++++++++++++----
 2 files changed, 43 insertions(+), 7 deletions(-)
meta/recipes-core/glibc/glibc.inc +3 -0
index 99a4753..4d5e98b 100644
@@ -15,6 +15,9 @@ def get_optimization(d):
     if bb.utils.contains("SELECTED_OPTIMIZATION", "-O0", "x", "", d) == "x":
         bb.note("glibc can't be built with -O0, -O2 will be used instead.")
         return selected_optimization.replace("-O0", "-O2")
+    elif bb.utils.contains("SELECTED_OPTIMIZATION", "-O0", "x", "", d) == "x":
+        bb.note("glibc can't be built with -Os, -Os -Wno-error will be used instead.")
+        return selected_optimization.replace("-Os", "-Os -Wno-error")
     return selected_optimization
 
 SELECTED_OPTIMIZATION := "${@get_optimization(d)}"
meta/recipes-core/glibc/glibc/eglibc-use-option-groups.patch +40 -7

91 lines not on this page · raw message

Signature · 1 line
2.1.4
#1 lore raw
Quoted 3 lines · Khem Raj
 .../glibc/glibc/eglibc-use-option-groups.patch     | 47

++++++++++++++++++----

Is this chunk related? The commit log only talks about disabling -Werror.

Ross

author
#2 lore raw
Quoted 9 lines · Ross

On 24 February 2015 at 06:39, Khem Raj <[email protected] <mailto:[email protected]>> wrote:

 .../glibc/glibc/eglibc-use-option-groups.patch     | 47 ++++++++++++++++++----

Is this chunk related? The commit log only talks about disabling -Werror.

yes it is. it fixed some of the obvious errors. but thats not enough. There is a lot more cleanup needed

Quoted 1 line

Ross

#3 lore raw
In reply to Khem Raj
Quoted 17 lines · Khem Raj

When we modify to use -Os
-Werror doesnt go well with it, glibc needs to be cleaned up for that but until then lets disable -Werror when using -Os

Change-Id I5495255fce67953f15c07e423e3e0eef41d4ce5e
---
  meta/recipes-core/glibc/glibc.inc                  |  3 ++
  .../glibc/glibc/eglibc-use-option-groups.patch     | 47 ++++++++++++++++++----
  2 files changed, 43 insertions(+), 7 deletions(-)
meta/recipes-core/glibc/glibc.inc +0 -0
index 99a4753..4d5e98b 100644

meta/recipes-core/glibc/glibc.inc def get_optimization(d): · line 20

Quoted 6 lines
meta/recipes-core/glibc/glibc.inc +3 -0
@@ -15,6 +15,9 @@ def get_optimization(d):
      if bb.utils.contains("SELECTED_OPTIMIZATION", "-O0", "x", "", d) == "x":
          bb.note("glibc can't be built with -O0, -O2 will be used instead.")
          return selected_optimization.replace("-O0", "-O2")
+    elif bb.utils.contains("SELECTED_OPTIMIZATION", "-O0", "x", "", d) == "x":
+        bb.note("glibc can't be built with -Os, -Os -Wno-error will be used instead.")
+        return selected_optimization.replace("-Os", "-Os -Wno-error")

I think there's a typo here which prevents this from working as intended. The second check is done on "-O0" when it's supposed to be "-Os"

Cheers,

Peter

#4 lore raw
Quoted 2 lines · Peter Urbanec

I think there's a typo here which prevents this from working as intended. The second check is done on "-O0" when it's supposed to be "-Os"

Tested patch submitted.

author
#5 lore raw
Quoted 5 lines · Peter Urbanec
Quoted 3 lines · Peter Urbanec

I think there's a typo here which prevents this from working as intended. The second check is done on "-O0" when it's supposed to be

"-Os"

Quoted 4 lines

Tested patch submitted.

Thanks

#6 lore raw
In reply to Khem Raj

Hello,

I get errors when:
MACHINE = "qemux86"
DEBUG_BUILD = "1"

$ bitbake glibc

../include/inline-hashtab.h: In function '_dl_make_tlsdesc_dynamic':
./tlsdeschtab.h:28:1: error: inlining failed in call to 'hash_tlsdesc': indirect function call with a yet undetermined callee [-Werror=inline]
hash_tlsdesc (void *p)
^
In file included from ./tlsdeschtab.h:25:0,

                 from ../sysdeps/i386/tlsdesc.c:25:

../include/inline-hashtab.h:154:3: error: called from here [-Werror=inline]
*find_empty_slot_for_expand (htab, hash_fn (*p))
^
In file included from ../sysdeps/i386/tlsdesc.c:25:0:
./tlsdeschtab.h:28:1: error: inlining failed in call to 'hash_tlsdesc': indirect function call with a yet undetermined callee [-Werror=inline]
hash_tlsdesc (void *p)
^
In file included from ./tlsdeschtab.h:25:0,

                 from ../sysdeps/i386/tlsdesc.c:25:
../include/inline-hashtab.h:201:8: error: called from here [-Werror=inline]
   hash = hash_fn (ptr);
        ^

In file included from ../sysdeps/i386/tlsdesc.c:25:0:
./tlsdeschtab.h:38:1: error: inlining failed in call to 'eq_tlsdesc': indirect function call with a yet undetermined callee [-Werror=inline]
eq_tlsdesc (void *p, void *q)
^
In file included from ./tlsdeschtab.h:25:0,

                 from ../sysdeps/i386/tlsdesc.c:25:
../include/inline-hashtab.h:209:12: error: called from here [-Werror=inline]
   else if (eq_fn (*entry, ptr))
            ^

In file included from ../sysdeps/i386/tlsdesc.c:25:0:
./tlsdeschtab.h:38:1: error: inlining failed in call to 'eq_tlsdesc': indirect function call with a yet undetermined callee [-Werror=inline]
eq_tlsdesc (void *p, void *q)
^
In file included from ./tlsdeschtab.h:25:0,

                 from ../sysdeps/i386/tlsdesc.c:25:

../include/inline-hashtab.h:222:16: error: called from here [-Werror=inline]
else if (eq_fn (*entry, ptr))

// Robert

Quoted 128 lines · Khem Raj

When we modify to use -Os
-Werror doesnt go well with it, glibc needs to be cleaned up for that but until then lets disable -Werror when using -Os

Change-Id I5495255fce67953f15c07e423e3e0eef41d4ce5e
---
  meta/recipes-core/glibc/glibc.inc                  |  3 ++
  .../glibc/glibc/eglibc-use-option-groups.patch     | 47 ++++++++++++++++++----
  2 files changed, 43 insertions(+), 7 deletions(-)
meta/recipes-core/glibc/glibc.inc +3 -0
index 99a4753..4d5e98b 100644
@@ -15,6 +15,9 @@ def get_optimization(d):
      if bb.utils.contains("SELECTED_OPTIMIZATION", "-O0", "x", "", d) == "x":
          bb.note("glibc can't be built with -O0, -O2 will be used instead.")
          return selected_optimization.replace("-O0", "-O2")
+    elif bb.utils.contains("SELECTED_OPTIMIZATION", "-O0", "x", "", d) == "x":
+        bb.note("glibc can't be built with -Os, -Os -Wno-error will be used instead.")
+        return selected_optimization.replace("-Os", "-Os -Wno-error")
      return selected_optimization
 
  SELECTED_OPTIMIZATION := "${@get_optimization(d)}"
meta/recipes-core/glibc/glibc/eglibc-use-option-groups.patch +40 -7
index 0fcaff9..439b4a9 100644
@@ -2294,11 +2294,37 @@ Index: git/locale/programs/locfile.h
   /* Change the output to be big-endian if BIG_ENDIAN is true and
      little-endian otherwise.  */
   static inline void
-@@ -276,4 +278,49 @@ extern void identification_output (struc
+@@ -89,7 +91,8 @@ maybe_swap_uint32 (uint32_t value)
+ }
+
+ /* Likewise, but munge an array of N uint32_ts starting at ARRAY.  */
+-static inline void
++static void
++__attribute__ ((unused))
+ maybe_swap_uint32_array (uint32_t *array, size_t n)
+ {
+   if (swap_endianness_p)
+@@ -99,7 +102,8 @@ maybe_swap_uint32_array (uint32_t *array
+
+ /* Like maybe_swap_uint32_array, but the array of N elements is at
+    the end of OBSTACK's current object.  */
+-static inline void
++static void
++__attribute__ ((unused))
+ maybe_swap_uint32_obstack (struct obstack *obstack, size_t n)
+ {
+   maybe_swap_uint32_array ((uint32_t *) obstack_next_free (obstack) - n, n);
+@@ -276,4 +280,55 @@ extern void identification_output (struc
   				   const struct charmap_t *charmap,
   				   const char *output_path);
 
-+static inline size_t
++static size_t wcslen_uint32 (const uint32_t *str) __attribute__ ((unused));
++static uint32_t * wmemcpy_uint32 (uint32_t *s1, const uint32_t *s2, size_t n) __attribute__ ((unused));
++static uint32_t * wcschr_uint32 (const uint32_t *s, uint32_t ch) __attribute__ ((unused));
++static int wcscmp_uint32 (const uint32_t *s1, const uint32_t *s2) __attribute__ ((unused));
++static int wmemcmp_uint32 (const uint32_t *s1, const uint32_t *s2, size_t n) __attribute__ ((unused));
++
++static size_t
  +wcslen_uint32 (const uint32_t *str)
  +{
  +  size_t len = 0;
@@ -2307,7 +2333,7 @@ Index: git/locale/programs/locfile.h
  +  return len;
  +}
  +
-+static inline int
++static  int
  +wmemcmp_uint32 (const uint32_t *s1, const uint32_t *s2, size_t n)
  +{
  +  while (n-- != 0)
@@ -2319,7 +2345,7 @@ Index: git/locale/programs/locfile.h
  +  return 0;
  +}
  +
-+static inline int
++static int
  +wcscmp_uint32 (const uint32_t *s1, const uint32_t *s2)
  +{
  +  while (*s1 != 0 && *s1 == *s2)
@@ -2327,13 +2353,13 @@ Index: git/locale/programs/locfile.h
  +  return *s1 - *s2;
  +}
  +
-+static inline uint32_t *
++static uint32_t *
  +wmemcpy_uint32 (uint32_t *s1, const uint32_t *s2, size_t n)
  +{
  +  return memcpy (s1, s2, n * sizeof (uint32_t));
  +}
  +
-+static inline uint32_t *
++static uint32_t *
  +wcschr_uint32 (const uint32_t *s, uint32_t ch)
  +{
  +  do
@@ -5627,7 +5653,7 @@ Index: git/posix/regexec.c
         /* match with character_class?  */
         for (i = 0; i < cset->nchar_classes; ++i)
   	{
-@@ -3863,8 +3843,16 @@ check_node_accept_bytes (const re_dfa_t
+@@ -3863,14 +3843,22 @@ check_node_accept_bytes (const re_dfa_t
   	      goto check_node_accept_bytes_match;
   	    }
   	}
@@ -5645,6 +5671,13 @@ Index: git/posix/regexec.c
         nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
         if (nrules != 0)
   	{
+ 	  unsigned int in_collseq = 0;
+ 	  const int32_t *table, *indirect;
+-	  const unsigned char *weights, *extra;
++	  const unsigned char *weights, *extra = NULL;
+ 	  const char *collseqwc;
+
+ 	  /* match with collating_symbol?  */
  @@ -3955,8 +3943,12 @@ check_node_accept_bytes (const re_dfa_t
   	    }
   	}
author
#7 lore raw
Quoted 11 lines · Robert Yang

Hello,

I get errors when:
MACHINE = "qemux86"
DEBUG_BUILD = "1"

$ bitbake glibc

We use -O ( which is -O1) when we do DEBUG builds. So same logic as -Os needs to go for -O as well. I will send a patch over weekend if no one does it before that