[Bug gcov-profile/55734] New: gcov-io.c uses builtins not available in non-GCC compilers

[Bug gcov-profile/55734] New: gcov-io.c uses builtins not available in non-GCC compilers

[aldyh at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

             Bug #: 55734
           Summary: gcov-io.c uses builtins not available in non-GCC
                    compilers
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: blocker
          Priority: P3
         Component: gcov-profile
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: aldyh@gcc.gnu.org


gcov-io.c uses the following builtins that are not available in non-GCC
compilers or in older GCC's (pre 3.4):

__builtin_popcountll
__builtin_clzll

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |build
   Last reconfirmed|                            |2012-12-18
     Ever Confirmed|0                           |1
            Summary|gcov-io.c uses builtins not |[4.8 Regression] gcov-io.c
                   |available in non-GCC        |uses builtins not available
                   |compilers                   |in non-GCC compilers
   Target Milestone|---                         |4.8.0

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-18 15:37:47 UTC ---
Confirmed.

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[aldyh at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #2 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2012-12-18 15:45:17 UTC ---
I have a proposed patch here:

http://gcc.gnu.org/ml/gcc-patches/2012-12/msg00766.html

...which was deemed not sexy at all.  So perhaps Richard or someone else can
suggest another route?

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #3 from Teresa Johnson <tejohnson at google dot com> 2012-12-18 15:49:03 UTC ---
In that thread, I had asked:

---
If you prefer, I can simply inline the popcount/clz functionality into
gcov-io.c directly (or at least when not using recent versions of
GCC). But that in fact would be duplicating the code, when I thought
Aldy's solution was trying to avoid that by providing the more general
interfaces.

Teresa
---

but didn't get a response. Richard, is that your preferred route for
fixing this issue?

Thanks,
Teresa

On Tue, Dec 18, 2012 at 7:45 AM, aldyh at gcc dot gnu.org
<gcc-bugzilla@gcc.gnu.org> wrote:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734
>
> --- Comment #2 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2012-12-18 15:45:17 UTC ---
> I have a proposed patch here:
>
> http://gcc.gnu.org/ml/gcc-patches/2012-12/msg00766.html
>
> ...which was deemed not sexy at all.  So perhaps Richard or someone else can
> suggest another route?
>
> --
> Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are on the CC list for the bug.

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-18 15:52:40 UTC ---
(In reply to comment #3)
> In that thread, I had asked:
> 
> ---
> If you prefer, I can simply inline the popcount/clz functionality into
> gcov-io.c directly (or at least when not using recent versions of
> GCC). But that in fact would be duplicating the code, when I thought
> Aldy's solution was trying to avoid that by providing the more general
> interfaces.
> 
> Teresa
> ---
> 
> but didn't get a response. Richard, is that your preferred route for
> fixing this issue?

We already have popcount_hwi and clz_hwi available.  Can't you piggy-back
on that to provide (inline in gcov-io.c) popcountll and clzll?  Why do
you need the long long variants anyway?

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-18 15:53:42 UTC ---
(In reply to comment #4)
> (In reply to comment #3)
> > In that thread, I had asked:
> > 
> > ---
> > If you prefer, I can simply inline the popcount/clz functionality into
> > gcov-io.c directly (or at least when not using recent versions of
> > GCC). But that in fact would be duplicating the code, when I thought
> > Aldy's solution was trying to avoid that by providing the more general
> > interfaces.
> > 
> > Teresa
> > ---
> > 
> > but didn't get a response. Richard, is that your preferred route for
> > fixing this issue?
> 
> We already have popcount_hwi and clz_hwi available.  Can't you piggy-back
> on that to provide (inline in gcov-io.c) popcountll and clzll?  Why do
> you need the long long variants anyway?

You don't seem to:

  unsigned histo_bitvector[GCOV_HISTOGRAM_BITVECTOR_SIZE];
...
         h_cnt += __builtin_popcountll (histo_bitvector[bv_ix]);

so just use popcount_hwi.

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[aldyh at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #6 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2012-12-18 16:19:33 UTC ---
Created attachment 28999
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28999
proposed patch

Oh, I didn't realize we always had the HWI variants available (inlined in
hwint.h).

Is something like this what you have in mind Richard?

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #7 from Teresa Johnson <tejohnson at google dot com> 2012-12-18 16:24:13 UTC ---
On Tue, Dec 18, 2012 at 7:53 AM, rguenth at gcc dot gnu.org
<gcc-bugzilla@gcc.gnu.org> wrote:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734
>
> --- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-18 15:53:42 UTC ---
> (In reply to comment #4)
>> (In reply to comment #3)
>> > In that thread, I had asked:
>> >
>> > ---
>> > If you prefer, I can simply inline the popcount/clz functionality into
>> > gcov-io.c directly (or at least when not using recent versions of
>> > GCC). But that in fact would be duplicating the code, when I thought
>> > Aldy's solution was trying to avoid that by providing the more general
>> > interfaces.
>> >
>> > Teresa
>> > ---
>> >
>> > but didn't get a response. Richard, is that your preferred route for
>> > fixing this issue?
>>
>> We already have popcount_hwi and clz_hwi available.  Can't you piggy-back
>> on that to provide (inline in gcov-io.c) popcountll and clzll?  Why do
>> you need the long long variants anyway?
>
> You don't seem to:
>
>   unsigned histo_bitvector[GCOV_HISTOGRAM_BITVECTOR_SIZE];
> ...
>          h_cnt += __builtin_popcountll (histo_bitvector[bv_ix]);
>
> so just use popcount_hwi.

The complication seems to be that hwint is not currently available for
libgcov.a (it is included by system.h which is not in libgcov.a). I
thought this was why Aldy took the approach of splitting out the hwi
routines into a separate helper file included by both hwint.c and
gcov-io. I can see if there is a way to simplify this, but I'm not
sure it is simple to include hwint into the gcov-io.c (and therefore
libgcov.a).

Teresa

>
> --
> Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are on the CC list for the bug.



--
Teresa Johnson | Software Engineer | tejohnson@google.com | 408-460-2413

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[aldyh at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

Aldy Hernandez <aldyh at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #28999|0                           |1
        is obsolete|                            |

--- Comment #8 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2012-12-18 16:25:49 UTC ---
Created attachment 29000
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29000
proposed patch

Duh, they're always available.  Like this?

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #9 from Teresa Johnson <tejohnson at google dot com> 2012-12-18 16:31:08 UTC ---
On Tue, Dec 18, 2012 at 8:25 AM, aldyh at gcc dot gnu.org
<gcc-bugzilla@gcc.gnu.org> wrote:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734
>
> Aldy Hernandez <aldyh at gcc dot gnu.org> changed:
>
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>   Attachment #28999|0                           |1
>         is obsolete|                            |
>
> --- Comment #8 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2012-12-18 16:25:49 UTC ---
> Created attachment 29000
>   --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29000
> proposed patch
>
> Duh, they're always available.  Like this?

Things will build, but you will presumably get an undefined symbol in
your libgcov.a, which would result in an undefined trying to build
with -fprofile-generate. Can you verify that this works. I just
checked with a similar change, which yields:

$ nm gcc/libgcov.a | grep popc
                 U popcount_hwi

/home/tejohnson/extra/gcc_trunk_5_validate/bld-gcc/gcc/libgcov.a(_gcov.o):
In function `__gcov_read_summary':
libgcov.c:(.text+0x7f3): undefined reference to `popcount_hwi'
collect2: error: ld returned 1 exit status

Teresa

>
> --
> Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are on the CC list for the bug.



--
Teresa Johnson | Software Engineer | tejohnson@google.com | 408-460-2413

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[aldyh at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #10 from Aldy Hernandez <aldyh at redhat dot com> 2012-12-18 16:48:40 UTC ---
Ah yes, now I remember.  Yes, there is a problem with libgcov.a.  I 
wasn't seeing it because I was only building cc1.  You are correct 
Teresa, that is the reason for the gymnastics in my original patch.

houston:/build/trunkboot/gcc$ nm libgcov.a |grep popc
                  U popcount_hwi

I will let you and Richard sort it out :).

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[aldyh at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #11 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2012-12-18 22:41:13 UTC ---
Teresa, I think this is what you were referring to, and the reason for my
original patch:

houston:/build/trunkboot$ ./install/bin/gcc a.c -fprofile-generate
/home/build/trunkboot/install/bin/../lib/gcc/x86_64-unknown-linux-gnu/4.8.0/libgcov.a(_gcov.o):
In function `gcov_histo_index':
libgcov.c:(.text+0x213): undefined reference to `clz_hwi'
/home/build/trunkboot/install/bin/../lib/gcc/x86_64-unknown-linux-gnu/4.8.0/libgcov.a(_gcov.o):
In function `__gcov_read_summary':
libgcov.c:(.text+0x7f3): undefined reference to `popcount_hwi'
collect2: error: ld returned 1 exit status

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #12 from Teresa Johnson <tejohnson at google dot com> 2012-12-18 22:49:20 UTC ---
On Tue, Dec 18, 2012 at 2:41 PM, aldyh at gcc dot gnu.org <
gcc-bugzilla@gcc.gnu.org> wrote:

>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734
>
> --- Comment #11 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2012-12-18
> 22:41:13 UTC ---
> Teresa, I think this is what you were referring to, and the reason for my
> original patch:
>
> houston:/build/trunkboot$ ./install/bin/gcc a.c -fprofile-generate
>
> /home/build/trunkboot/install/bin/../lib/gcc/x86_64-unknown-linux-gnu/4.8.0/libgcov.a(_gcov.o):
> In function `gcov_histo_index':
> libgcov.c:(.text+0x213): undefined reference to `clz_hwi'
>
> /home/build/trunkboot/install/bin/../lib/gcc/x86_64-unknown-linux-gnu/4.8.0/libgcov.a(_gcov.o):
> In function `__gcov_read_summary':
> libgcov.c:(.text+0x7f3): undefined reference to `popcount_hwi'
> collect2: error: ld returned 1 exit status
>

Yep. I've got a patch in the works that is a slightly simplified version of
what you did. I'll send that a little later today once I get it tested.

Teresa

>
> --
> Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are on the CC list for the bug.
>

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #13 from Teresa Johnson <tejohnson at google dot com> 2012-12-19 06:49:50 UTC ---
On Tue, Dec 18, 2012 at 2:48 PM, Teresa Johnson <tejohnson@google.com> wrote:
>
>
>
>
> On Tue, Dec 18, 2012 at 2:41 PM, aldyh at gcc dot gnu.org <gcc-bugzilla@gcc.gnu.org> wrote:
>>
>>
>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734
>>
>> --- Comment #11 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2012-12-18 22:41:13 UTC ---
>> Teresa, I think this is what you were referring to, and the reason for my
>> original patch:
>>
>> houston:/build/trunkboot$ ./install/bin/gcc a.c -fprofile-generate
>> /home/build/trunkboot/install/bin/../lib/gcc/x86_64-unknown-linux-gnu/4.8.0/libgcov.a(_gcov.o):
>> In function `gcov_histo_index':
>> libgcov.c:(.text+0x213): undefined reference to `clz_hwi'
>> /home/build/trunkboot/install/bin/../lib/gcc/x86_64-unknown-linux-gnu/4.8.0/libgcov.a(_gcov.o):
>> In function `__gcov_read_summary':
>> libgcov.c:(.text+0x7f3): undefined reference to `popcount_hwi'
>> collect2: error: ld returned 1 exit status
>
>
> Yep. I've got a patch in the works that is a slightly simplified version of what you did. I'll send that a little later today once I get it tested.


Here is the patch. I am doing some final regression testing on it, but
earlier testing and my own testing looked good. It is similar to what
Aldy did, pulling some of the hwi routines into a new hwint-helper.c
file. However, in this case, it is included normally into hwint.c, and
the routines are only pulled in as static copies to gcov-io.c when we
are building libgcov.a (in other cases gcov-io.c is being built when
system.h is included, which includes hwint.h already). In the case of
libgcov.a, I also had to add support for the fact that the HOST_*
variables that were being used in the hwint routines are not defined.

Richard, let me know if this looks like a reasonable approach. The
only other solution I had was to simply inline copies of popcount_hwi
and clz_hwi into gcov-io.c when we are building libgcov.a.

Thanks,
Teresa

2012-12-18  Teresa Johnson  <tejohnson@google.com>
            Aldy Hernandez  <aldyh@redhat.com>

* hwint-helper.c (floor_log2): Moved from hwint.c
(ceil_log2): Ditto.
(exact_log2): Ditto.
(ctz_hwi): Ditto.
(clz_hwi): Ditto.
(ffs_hwi): Ditto.
(popcount_hwi): Ditto.
* gcov-io.c (gcov_read_summary): Invoke popcount_hwi.
(gcov_histo_index): Invoke clz_hwi.
* Makefile.in (hwint.o): Depend on hwint-helper.c
* hwint.c (floor_log2): Moved to hwint-helper.c
(ceil_log2): Ditto.
(exact_log2): Ditto.
(ctz_hwi): Ditto.
(clz_hwi): Ditto.
(ffs_hwi): Ditto.
(popcount_hwi): Ditto.

Index: hwint-helper.c
===================================================================
--- hwint-helper.c (revision 0)
+++ hwint-helper.c (revision 0)
@@ -0,0 +1,130 @@
+/* Supporting bit functions for older versions of GCC, and non-GCC
+   bootstrap compilers.
+
+   Copyright (C) 1987, 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
+   1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010,
+   2012 Free Software Foundation, Inc.
+
+This file is part of GCC.
+
+GCC is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 3, or (at your option) any later
+version.
+
+GCC is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
+<http://www.gnu.org/licenses/>.  */
+
+#if GCC_VERSION < 3004
+
+/* The functions clz_hwi, ctz_hwi, ffs_hwi, floor_log2, ceil_log2,
+   and exact_log2 are defined as inline functions in hwint.h
+   if GCC_VERSION >= 3004.
+   The definitions here are used for older versions of GCC and
+   non-GCC bootstrap compilers.  */
+
+/* Given X, an unsigned number, return the largest int Y such that 2**Y <= X.
+   If X is 0, return -1.  */
+
+HWINT_LINKAGE int
+floor_log2 (HWINT_TYPE x)
+{
+  int t = 0;
+
+  if (x == 0)
+    return -1;
+
+  if (HWINT_TYPE_SIZE > 64)
+    if (x >= (HWINT_TYPE) 1 << (t + 64))
+      t += 64;
+  if (HWINT_TYPE_SIZE > 32)
+    if (x >= ((HWINT_TYPE) 1) << (t + 32))
+      t += 32;
+  if (x >= ((HWINT_TYPE) 1) << (t + 16))
+    t += 16;
+  if (x >= ((HWINT_TYPE) 1) << (t + 8))
+    t += 8;
+  if (x >= ((HWINT_TYPE) 1) << (t + 4))
+    t += 4;
+  if (x >= ((HWINT_TYPE) 1) << (t + 2))
+    t += 2;
+  if (x >= ((HWINT_TYPE) 1) << (t + 1))
+    t += 1;
+
+  return t;
+}
+
+/* Given X, an unsigned number, return the largest Y such that 2**Y >= X.  */
+
+HWINT_LINKAGE int
+ceil_log2 (HWINT_TYPE x)
+{
+  return floor_log2 (x - 1) + 1;
+}
+
+/* Return the logarithm of X, base 2, considering X unsigned,
+   if X is a power of 2.  Otherwise, returns -1.  */
+
+HWINT_LINKAGE int
+exact_log2 (HWINT_TYPE x)
+{
+  if (x != (x & -x))
+    return -1;
+  return floor_log2 (x);
+}
+
+/* Given X, an unsigned number, return the number of least significant bits
+   that are zero.  When X == 0, the result is the word size.  */
+
+HWINT_LINKAGE int
+ctz_hwi (HWINT_TYPE x)
+{
+  return x ? floor_log2 (x & -x) : HWINT_TYPE_SIZE;
+}
+
+/* Similarly for most significant bits.  */
+
+HWINT_LINKAGE int
+clz_hwi (HWINT_TYPE x)
+{
+  return HWINT_TYPE_SIZE - 1 - floor_log2 (x);
+}
+
+/* Similar to ctz_hwi, except that the least significant bit is numbered
+   starting from 1, and X == 0 yields 0.  */
+
+HWINT_LINKAGE int
+ffs_hwi (HWINT_TYPE x)
+{
+  return 1 + floor_log2 (x & -x);
+}
+
+/* Return the number of set bits in X.  */
+
+HWINT_LINKAGE int
+popcount_hwi (HWINT_TYPE x)
+{
+  int i, ret = 0;
+  size_t bits = sizeof (x) * CHAR_BIT;
+
+  for (i = 0; i < (int)bits; i += 1)
+    {
+      ret += x & 1;
+      x >>= 1;
+    }
+
+  return ret;
+}
+
+#endif /* GCC_VERSION < 3004 */
Index: gcov-io.c
===================================================================
--- gcov-io.c (revision 194585)
+++ gcov-io.c (working copy)
@@ -517,6 +517,18 @@ gcov_read_string (void)
 }
 #endif

+#if IN_LIBGCOV
+#if GCC_VERSION < 3004
+#define HWINT_LINKAGE inline static
+#define HWINT_TYPE gcov_type_unsigned
+#define HWINT_TYPE_SIZE LONG_LONG_TYPE_SIZE
+#include "hwint-helper.c"
+#else
+#define popcount_hwi __builtin_popcountll
+#define clz_hwi __builtin_clzll
+#endif
+#endif
+
 GCOV_LINKAGE void
 gcov_read_summary (struct gcov_summary *summary)
 {
@@ -538,7 +550,7 @@ gcov_read_summary (struct gcov_summary *summary)
       for (bv_ix = 0; bv_ix < GCOV_HISTOGRAM_BITVECTOR_SIZE; bv_ix++)
         {
           histo_bitvector[bv_ix] = gcov_read_unsigned ();
-          h_cnt += __builtin_popcountll (histo_bitvector[bv_ix]);
+          h_cnt += popcount_hwi (histo_bitvector[bv_ix]);
         }
       bv_ix = 0;
       h_ix = 0;
@@ -642,7 +654,7 @@ gcov_histo_index (gcov_type value)

   /* Find the place of the most-significant bit set.  */
   if (v > 0)
-    r = 63 - __builtin_clzll (v);
+    r = 63 - clz_hwi (v);

   /* If at most the 2 least significant bits are set (value is
      0 - 3) then that value is our index into the lowest set of
Index: Makefile.in
===================================================================
--- Makefile.in (revision 194585)
+++ Makefile.in (working copy)
@@ -2701,7 +2701,7 @@ toplev.o : toplev.c $(CONFIG_H) $(SYSTEM_H) corety
    $(TREE_PRETTY_PRINT_H) opts-diagnostic.h $(COMMON_TARGET_H) \
    tsan.h

-hwint.o : hwint.c $(CONFIG_H) $(SYSTEM_H) $(DIAGNOSTIC_CORE_H)
+hwint.o : hwint.c $(CONFIG_H) $(SYSTEM_H) $(DIAGNOSTIC_CORE_H) hwint-helper.c

 passes.o : passes.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(TREE_H) \
    $(RTL_H) $(FUNCTION_H) $(FLAGS_H) $(INPUT_H) $(INSN_ATTR_H) output.h \
Index: hwint.c
===================================================================
--- hwint.c (revision 194585)
+++ hwint.c (working copy)
@@ -23,109 +23,11 @@ along with GCC; see the file COPYING3.  If not see
 #include "system.h"
 #include "diagnostic-core.h"

-#if GCC_VERSION < 3004
+#define HWINT_LINKAGE
+#define HWINT_TYPE unsigned HOST_WIDE_INT
+#define HWINT_TYPE_SIZE HOST_BITS_PER_WIDE_INT
+#include "hwint-helper.c"

-/* The functions clz_hwi, ctz_hwi, ffs_hwi, floor_log2, ceil_log2,
-   and exact_log2 are defined as inline functions in hwint.h
-   if GCC_VERSION >= 3004.
-   The definitions here are used for older versions of GCC and
-   non-GCC bootstrap compilers.  */
-
-/* Given X, an unsigned number, return the largest int Y such that 2**Y <= X.
-   If X is 0, return -1.  */
-
-int
-floor_log2 (unsigned HOST_WIDE_INT x)
-{
-  int t = 0;
-
-  if (x == 0)
-    return -1;
-
-  if (HOST_BITS_PER_WIDE_INT > 64)
-    if (x >= (unsigned HOST_WIDE_INT) 1 << (t + 64))
-      t += 64;
-  if (HOST_BITS_PER_WIDE_INT > 32)
-    if (x >= ((unsigned HOST_WIDE_INT) 1) << (t + 32))
-      t += 32;
-  if (x >= ((unsigned HOST_WIDE_INT) 1) << (t + 16))
-    t += 16;
-  if (x >= ((unsigned HOST_WIDE_INT) 1) << (t + 8))
-    t += 8;
-  if (x >= ((unsigned HOST_WIDE_INT) 1) << (t + 4))
-    t += 4;
-  if (x >= ((unsigned HOST_WIDE_INT) 1) << (t + 2))
-    t += 2;
-  if (x >= ((unsigned HOST_WIDE_INT) 1) << (t + 1))
-    t += 1;
-
-  return t;
-}
-
-/* Given X, an unsigned number, return the largest Y such that 2**Y >= X.  */
-
-int
-ceil_log2 (unsigned HOST_WIDE_INT x)
-{
-  return floor_log2 (x - 1) + 1;
-}
-
-/* Return the logarithm of X, base 2, considering X unsigned,
-   if X is a power of 2.  Otherwise, returns -1.  */
-
-int
-exact_log2 (unsigned HOST_WIDE_INT x)
-{
-  if (x != (x & -x))
-    return -1;
-  return floor_log2 (x);
-}
-
-/* Given X, an unsigned number, return the number of least significant bits
-   that are zero.  When X == 0, the result is the word size.  */
-
-int
-ctz_hwi (unsigned HOST_WIDE_INT x)
-{
-  return x ? floor_log2 (x & -x) : HOST_BITS_PER_WIDE_INT;
-}
-
-/* Similarly for most significant bits.  */
-
-int
-clz_hwi (unsigned HOST_WIDE_INT x)
-{
-  return HOST_BITS_PER_WIDE_INT - 1 - floor_log2(x);
-}
-
-/* Similar to ctz_hwi, except that the least significant bit is numbered
-   starting from 1, and X == 0 yields 0.  */
-
-int
-ffs_hwi (unsigned HOST_WIDE_INT x)
-{
-  return 1 + floor_log2 (x & -x);
-}
-
-/* Return the number of set bits in X.  */
-
-int
-popcount_hwi (unsigned HOST_WIDE_INT x)
-{
-  int i, ret = 0;
-  size_t bits = sizeof (x) * CHAR_BIT;
-
-  for (i = 0; i < bits; i += 1)
-    {
-      ret += x & 1;
-      x >>= 1;
-    }
-
-  return ret;
-}
-
-#endif /* GCC_VERSION < 3004 */
-
 /* Compute the absolute value of X.  */

 HOST_WIDE_INT

>
> Teresa
>>
>>
>> --
>> Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
>> ------- You are receiving this mail because: -------
>> You are on the CC list for the bug.
>
>
>
>
> --
> Teresa Johnson | Software Engineer |  tejohnson@google.com |  408-460-2413
>



--
Teresa Johnson | Software Engineer |  tejohnson@google.com |  408-460-2413

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #14 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-19 09:43:27 UTC ---
(In reply to comment #10)
> Ah yes, now I remember.  Yes, there is a problem with libgcov.a.  I 
> wasn't seeing it because I was only building cc1.  You are correct 
> Teresa, that is the reason for the gymnastics in my original patch.
> 
> houston:/build/trunkboot/gcc$ nm libgcov.a |grep popc
>                   U popcount_hwi
> 
> I will let you and Richard sort it out :).

libgcov should be built by stage1 gcc already and thus has the builtins
available.

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-19 13:07:01 UTC ---
(In reply to comment #14)
> (In reply to comment #10)
> > Ah yes, now I remember.  Yes, there is a problem with libgcov.a.  I 
> > wasn't seeing it because I was only building cc1.  You are correct 
> > Teresa, that is the reason for the gymnastics in my original patch.
> > 
> > houston:/build/trunkboot/gcc$ nm libgcov.a |grep popc
> >                   U popcount_hwi
> > 
> > I will let you and Richard sort it out :).
> 
> libgcov should be built by stage1 gcc already and thus has the builtins
> available.

To be more explicit - use #if IN_LIBGCOV to guard __builtin_popcountll vs.
popcount_hwi.

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #16 from Teresa Johnson <tejohnson at google dot com> 2012-12-19 15:07:54 UTC ---
On Wed, Dec 19, 2012 at 5:07 AM, rguenth at gcc dot gnu.org
<gcc-bugzilla@gcc.gnu.org> wrote:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734
>
> --- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-19 13:07:01 UTC ---
> (In reply to comment #14)
>> (In reply to comment #10)
>> > Ah yes, now I remember.  Yes, there is a problem with libgcov.a.  I
>> > wasn't seeing it because I was only building cc1.  You are correct
>> > Teresa, that is the reason for the gymnastics in my original patch.
>> >
>> > houston:/build/trunkboot/gcc$ nm libgcov.a |grep popc
>> >                   U popcount_hwi
>> >
>> > I will let you and Richard sort it out :).
>>
>> libgcov should be built by stage1 gcc already and thus has the builtins
>> available.
>
> To be more explicit - use #if IN_LIBGCOV to guard __builtin_popcountll vs.
> popcount_hwi.

Ah, yes, this fact escaped me and makes things much simpler. Running
the regression tests right now, will send patch as soon as they
finish...

Thanks,
Teresa

>
> --
> Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are on the CC list for the bug.



--
Teresa Johnson | Software Engineer | tejohnson@google.com | 408-460-2413

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #17 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-19 16:39:56 UTC ---
I guess a problem of such approach is if not in libgcov.a and HOST_WIDE_INT is
smaller than long long (or for clz even not exactly 64-bit).
For popcount, I wonder why the code uses __builtin_popcountll at all, when the
value is unsigned int, it should just use __builtin_popcount and I think we can
safely assume that HOST_WIDE_INT is at least as wide as int.
So just using either popcount_hwi or __builtin_popcount (no ll) should be fine.
For clz, there is:
  if (v > 0)
    r = 63 - __builtin_clzll (v);
i.e. it relies on clzll having exactly 64-bit argument.  For __builtin_clzll
perhaps the assumption is fine, for clz_hwi not so.  So you need some extra
code to implement clz_ll on top of clz_hwi if HWI is smaller than 64-bit.

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #18 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-19 16:48:00 UTC ---
Perhaps
  if (v > 0)
    {
#if IN_LIBGCOV
      r = 63 - __builtin_clzll (v);
#else
      if (sizeof (v) <= sizeof (HOST_WIDE_INT))
        r = floor_log2 (v);
      else if (sizeof (v) == 2 * sizeof (HOST_WIDE_INT))
        {
          r = v >> HOST_BITS_PER_WIDE_INT;
          if (r)
            r = floor_log2 (r) + HOST_BITS_PER_WIDE_INT;
          else
            r = floor_log2 (v);
        }
      else
        gcc_unreachable ();
#endif
    }

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #19 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-19 16:57:31 UTC ---
Of course v >> HOST_BITS_PER_WIDE_INT can't be assigned to r, it needs some new
temporary, with unsigned HOST_WIDE_INT or gcov_type_unsigned type (and use that
temporary instead in if (r) and floor_log2 (r).

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #20 from Teresa Johnson <tejohnson at google dot com> 2012-12-19 17:07:51 UTC ---
On Wed, Dec 19, 2012 at 8:39 AM, jakub at gcc dot gnu.org
<gcc-bugzilla@gcc.gnu.org> wrote:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734
>
> Jakub Jelinek <jakub at gcc dot gnu.org> changed:
>
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |jakub at gcc dot gnu.org
>
> --- Comment #17 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-19 16:39:56 UTC ---
> I guess a problem of such approach is if not in libgcov.a and HOST_WIDE_INT is
> smaller than long long (or for clz even not exactly 64-bit).
> For popcount, I wonder why the code uses __builtin_popcountll at all, when the
> value is unsigned int, it should just use __builtin_popcount and I think we can
> safely assume that HOST_WIDE_INT is at least as wide as int.
> So just using either popcount_hwi or __builtin_popcount (no ll) should be fine.

ok.

> For clz, there is:
>   if (v > 0)
>     r = 63 - __builtin_clzll (v);
> i.e. it relies on clzll having exactly 64-bit argument.  For __builtin_clzll
> perhaps the assumption is fine, for clz_hwi not so.  So you need some extra
> code to implement clz_ll on top of clz_hwi if HWI is smaller than 64-bit.

ok, thanks for pointing out this issue. I'll try out the solution you posted.

Teresa

>
> --
> Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are on the CC list for the bug.



--
Teresa Johnson | Software Engineer | tejohnson@google.com | 408-460-2413

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #21 from Teresa Johnson <tejohnson at google dot com> 2012-12-19 17:35:08 UTC ---
On Wed, Dec 19, 2012 at 8:48 AM, jakub at gcc dot gnu.org
<gcc-bugzilla@gcc.gnu.org> wrote:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734
>
> --- Comment #18 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-19 16:48:00 UTC ---
> Perhaps
>   if (v > 0)
>     {
> #if IN_LIBGCOV
>       r = 63 - __builtin_clzll (v);
> #else
>       if (sizeof (v) <= sizeof (HOST_WIDE_INT))
>         r = floor_log2 (v);
>       else if (sizeof (v) == 2 * sizeof (HOST_WIDE_INT))
>         {
>           r = v >> HOST_BITS_PER_WIDE_INT;
>           if (r)
>             r = floor_log2 (r) + HOST_BITS_PER_WIDE_INT;
>           else
>             r = floor_log2 (v);
>         }
>       else
>         gcc_unreachable ();
> #endif
>     }

Also, I have changed the runtime size checks into preprocessor checks,
both to increase the speed and also to avoid compile-time errors about
shifting by the size of the variable (on v >> HOST_BITS_PER_WIDE_INT).

Thanks,
Teresa

>
> --
> Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are on the CC list for the bug.



--
Teresa Johnson | Software Engineer | tejohnson@google.com | 408-460-2413

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[tejohnson at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

--- Comment #22 from tejohnson at gcc dot gnu.org 2012-12-20 14:31:18 UTC ---
Author: tejohnson
Date: Thu Dec 20 14:31:09 2012
New Revision: 194634

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=194634
Log:
Fix PR gcov-profile/55734 by using methods from hwint.c instead of
builtins, to handle non-GCC and older versions of GCC. When building
libgcov.a, however, hwint.c is not available, but we are always using
the bootstrapped compiler and can therefore use the builtins.

Use __builtin_popcount instead of __builtin_popcountll, since we
are operating on an int.

Use floor_log2 directly, instead of clz_hwi for the non-libgcov case,
and handle situations where the size of the gcov_type is bigger than
HOST_WIDE_INT. Verified that the various cases compiled by forcing
different HOST_BITS_PER_WIDE_INT values.

2012-12-20  Teresa Johnson  <tejohnson@google.com>
            Jakub Jelinek  <jakub@redhat.com>

        PR gcov-profile/55734
    * gcov-io.c (gcov_read_summary): Use __builtin_popcount instead
        of __builtin_popcountll when building libgcov.a, otherwise use
        popcount_hwi.
    (gcov_histo_index): When not building libgcov.a, use floor_log2
        instead of __builtin_clzll.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/gcov-io.c

[Bug gcov-profile/55734] [4.8 Regression] gcov-io.c uses builtins not available in non-GCC compilers

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55734

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #23 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-20 14:56:27 UTC ---
Fixed.