Re: [PATCH] Partial Transition fix attempt

On Wed, Mar 21, 2012 at 6:44 PM, redbrain wrote: > in tree-mudflap.c to change usage of fold_conver to build_int_cst > http://gcc.gnu.org/wiki/Partial_Transitions. I am not 100% sure this is > correct but maybe someone can shed some light. Doesn't look ok - how do you know you even have constan

Re: [PATCH][Testsuite] XFAIL scev-3/4.c and add scev-5.c

On Thu, Mar 22, 2012 at 3:19 AM, Jiangning Liu wrote: > Hi, > > This patch is to XFAIL scev-3.c and scev-5.c. > > The bug is going to be fixed after Richard Guenther fix a serials of > problems related to POINTER_PLUS_EXPR and sizetype precision. Ok. Thanks, Richard. > Thanks, > -Jiangning > >

[PATCH] Fix PR52459

This fixes PR52459, we re-use VN reference to encode calls but we are not interested in restricting call insertions in PREs inhibit_phi_insertion (we at most have pure calls). Boostrapped and tested on x86_64-unknown-linux-gnu, applied to trunk. Richard. 2012-03-21 Richard Guenther

Re: [PATCH, alpha]: Use generic config/elfos.h headers

On Wed, Mar 21, 2012 at 11:37 PM, Richard Henderson wrote: >> Attached patch enables alpha to use generic config/elfos.h headers on >> linux and *bsd targets. The most important difference to generic >> elfos.h is in >> >>       * config/alpha/elf.h (TARGET_ASM_FILE_START_FILE_DIRECTIVE): Undefin

Re: [RFC][PATCH] A change to do_while_loop_p()

On Wed, 21 Mar 2012, Razya Ladelsky wrote: > Hi, > > I need to use do_while_loop_p, but I'm not sure its functionality is what > I expected it to be. > > This is the part that I do not understand: > > /* If the header contains just a condition, it is not a do-while loop. */ > stmt = last_an

[PATCH, MIPS] Add -march=xlp

This patch adds basic support for Broadcom XLP (formerly NetLogic XLP) to GCC. Tested by building a mips64-linux-gnu toolchain with -march=xlp. OK to check in? -- Maxim Kuvyrkov CodeSourcery / Mentor Graphics gcc-basic-xlp.ChangeLog Description: Binary data gcc-basic-xlp.patch Description:

[Patch V2] libgfortran: do not assume libm

Hi, this is version 2 of the patch. The initial problem is that libgfortran configure.ac used AC_CHECK_LIB([m]…) to check wether several math functions are available. That doesn't work on VMS, because there is no such things as a libm. It seems to me that there are no autoconf macro to check

Re: [Patch V2] libgfortran: do not assume libm

Il 22/03/2012 09:30, Tristan Gingold ha scritto: > Hi, > > this is version 2 of the patch. > > The initial problem is that libgfortran configure.ac used AC_CHECK_LIB([m]…) > to check wether several math functions are available. That doesn't work on > VMS, because there is no such things as a l

Re: remove wrong code in immed_double_const

Mike Stump writes: >> This is changing the real case, where sign extension doesn't make >> much sense. > > I'm not sure I followed. Do you want me to remove the change for the > second case, leave it as it, or something else? I've left it as I had > modified it. Sorry, meant we should leave the

Re: [PATCH, MIPS] Add -march=xlp

Maxim Kuvyrkov writes: > This patch adds basic support for Broadcom XLP (formerly NetLogic XLP) to GCC. > > Tested by building a mips64-linux-gnu toolchain with -march=xlp. > > OK to check in? The new option needs to be documented in invoke.texi, so: @samp{vr5000}, @samp{vr5400}, @samp{vr550

Re: [C++ Patch] PR 49152

Hi, On 03/22/2012 05:25 AM, Gabriel Dos Reis wrote: On Wed, Mar 21, 2012 at 7:22 PM, Paolo Carlini wrote: Hi, this diagnostic issue is about not even trying to print expressions in error messages involving operators, and print operand types instead. Just as an example, for: struct X { int x;

Re: remove wrong code in immed_double_const

Richard Sandiford writes: > Should be: > > if (add_double_with_sign (l1, h1, l2, h2, &lv, &hv, false)) > /* Sorry, we have no way to represent overflows this >wide. To fix, add constant support wider than >CONST_DOUBLE. */ > gcc_assert (GET_MODE_BITS

Re: [PATCH] Fix PRs 52080, 52097 and 48124, rewrite bitfield expansion, enable the C++ memory model wrt bitfields everywhere

> This is what I have applied after bootstrapping and testing on > x86_64-unknown-linux-gnu. Thanks. Last identified problem: the miscompilation with variant record. gnat.dg/pack17.adb raised PROGRAM_ERROR : pack17.adb:36 explicit raise FAIL: gnat.dg/pack17.adb execution test Things start t

Re: [PATCH] Fix PRs 52080, 52097 and 48124, rewrite bitfield expansion, enable the C++ memory model wrt bitfields everywhere

On Thu, 22 Mar 2012, Eric Botcazou wrote: > > This is what I have applied after bootstrapping and testing on > > x86_64-unknown-linux-gnu. > > Thanks. Last identified problem: the miscompilation with variant record. > > gnat.dg/pack17.adb > > raised PROGRAM_ERROR : pack17.adb:36 explicit rai

[Backport,4.7,committed]: Fix PR other/52545: SECTION_EXCLUE clobbers SECTION_MACH_DEP

http://gcc.gnu.org/viewcvs?view=revision&revision=185685 Applied as obvious. Johann Backport from mainline r185259. PR other/52545 * output.h (SECTION_EXCLUDE, SECTION_MACH_DEP): Don't use SECTION_MACH_DEP reserved bits for SECTION_EXCLUDE.

Re: [PATCH] Fix PRs 52080, 52097 and 48124, rewrite bitfield expansion, enable the C++ memory model wrt bitfields everywhere

> Yes, I have noticed that from what you can see in the code. So the > issue is that get_bit_range tells us that it's ok to touch bits > outside of the field - but that's obviously required. We may not > change bits that are not covered by the FIELD_DECL which now > somehow happens? That sounds

Re: [PATCH] Fix PRs 52080, 52097 and 48124, rewrite bitfield expansion, enable the C++ memory model wrt bitfields everywhere

On Thu, 22 Mar 2012, Eric Botcazou wrote: > > Yes, I have noticed that from what you can see in the code. So the > > issue is that get_bit_range tells us that it's ok to touch bits > > outside of the field - but that's obviously required. We may not > > change bits that are not covered by the FI

[PATCH] Fix PR52638

This fixes PR52638, I made vect_init_vector handle scalar inits to not duplicate this code all over the place. Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk. Richard. 2012-03-22 Richard Guenther PR tree-optimization/52638 * tree-vect-stmts.c (vect_ini

Re: [PATCH] Fix PRs 52080, 52097 and 48124, rewrite bitfield expansion, enable the C++ memory model wrt bitfields everywhere

> But the bitfield group _does_ start on a byte boundary. At least > that's what the new code in stor-layout ensures. No, it doesn't, since it does the computation on a record by record basis. Here we have a bitfield group that starts in a record and ends up in a nested record. > It's ok to ass

Re: [PATCH] Fix PRs 52080, 52097 and 48124, rewrite bitfield expansion, enable the C++ memory model wrt bitfields everywhere

On Thu, Mar 22, 2012 at 12:47 PM, Eric Botcazou wrote: >> But the bitfield group _does_ start on a byte boundary.  At least >> that's what the new code in stor-layout ensures. > > No, it doesn't, since it does the computation on a record by record basis. > Here we have a bitfield group that starts

Re: [PATCH] Fix PRs 52080, 52097 and 48124, rewrite bitfield expansion, enable the C++ memory model wrt bitfields everywhere

> bitregion_start == 11 looks bogus. The representative is starting at > > DECL_FIELD_BIT_OFFSET (repr) > = size_binop (BIT_AND_EXPR, > DECL_FIELD_BIT_OFFSET (field), > bitsize_int (~(BITS_PER_UNIT - 1))); > > which looks ok It cannot be OK if you want it

Re: [PATCH] gfortran testsuite: implicitly cleanup-modules

On Mon, Mar 19, 2012 at 06:36:12PM +0100, Rainer Orth wrote: >Unfortunately, this patch creates a bogus warning in the libgomp >testsuite, which also shows up in mail-report.log: > >WARNING: Line 30 includes unreadable file >\`/vol/gcc/src/hg/trunk/local/libgomp/testsuite/libgomp.fortran/omp_lib.

Re: [PATCH] Fix PRs 52080, 52097 and 48124, rewrite bitfield expansion, enable the C++ memory model wrt bitfields everywhere

On Thu, 22 Mar 2012, Eric Botcazou wrote: > > bitregion_start == 11 looks bogus. The representative is starting at > > > > DECL_FIELD_BIT_OFFSET (repr) > > = size_binop (BIT_AND_EXPR, > > DECL_FIELD_BIT_OFFSET (field), > > bitsize_int (~(BITS_PER_UNIT - 1

[PATCH, alpha]: Some further configure cleanups

Hello! This patch moves a couple of #defines from alpha/alpha.h to alpha/vms.h to use generic defines from elfos.h. 2012-03-22 Uros Bizjak * config/alpha/alpha.h (ASM_OUTPUT_ALIGN): Move to elf.h (ASM_OUTPUT_SKIP): Move to vms.h (ASM_OUTPUT_LOCAL): Ditto. (ASM_

Re: [PATCH, alpha]: Some further configure cleanups

On Mar 22, 2012, at 1:49 PM, Uros Bizjak wrote: > Hello! > > This patch moves a couple of #defines from alpha/alpha.h to > alpha/vms.h to use generic defines from elfos.h. > > 2012-03-22 Uros Bizjak > > * config/alpha/alpha.h (ASM_OUTPUT_ALIGN): Move to elf.h > (ASM_OUTPUT_SKIP)

[PATCH] Fix PR52548

This fixes PR52548, we were not flow-sensitively pruning EXP_GEN sets during antic-in computation. Fixed by simply pruning EXP_GEN at generation time. Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk. Richard. 2012-03-22 Richard Guenther PR tree-optimization/52

Re: remove wrong code in immed_double_const

Hi, On Wed, 21 Mar 2012, Mike Stump wrote: > > If the high bit of i1 is set then currently you will get a negative > > number produced no matter the absolute value of it. > > Ok, in the new patch, I'm pushing to change the spec so that the value > is sign extended and fixing all the code that

Re: [Patch,AVR,committed]: Ad PR52496: Use unspec_volatile instead of unspec

Georg-Johann Lay wrote: > This patch adds memory barriers to > > __builtin_avr_nop > __builtin_avr_sei > __builtin_avr_cli > __builtin_avr_wdr > __builtin_avr_sleep > __builtin_avr_delay_cycles > > so that their code cannot be dragged over memory accesses. > > Ok for trunk? > > PR target/

[Patch] mpfr headers check

Hi, Attached is a rebased copy of a patch from . This is mostly a ping reminder. Index: configure.ac === --- configure.ac(revision 185692) +++ configure.ac(working co

[PATCH] Fix PR52666

Committed as obvious. Richard. 2012-03-22 Richard Guenther PR middle-end/52666 * fold-const.c (fold_ternary_loc): Fix typo. Index: gcc/fold-const.c === --- gcc/fold-const.c(revision 185691) +++ gcc/fold-cons

Re: remove wrong code in immed_double_const

Hi, On Wed, 21 Mar 2012, Mike Stump wrote: > --- a/gcc/emit-rtl.c > +++ b/gcc/emit-rtl.c > @@ -531,10 +531,9 @@ immed_double_const (HOST_WIDE_INT i0, HOST_WIDE_INT i1, > enum machine_mode mode) > > 1) If GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT, then we use > gen_int_mode.

Patch for -mno-exr option in case of function with "monitor" attribute

Hi, Please find the attached patch to avoid saving of EXR register for monitor functions. By default, in prologue code of a monitor function, EXR register is pushed onto the stack. This implementation is not required for H8S/224x and 21xx variants of H8S controllers. The behavior can be controlle

PR middle-end/51737

Hi, this patch fixes (now latent) problem where save_inline_function_body blows up when self recursive call is removed during the body duplication. The problem reproduce on 4.6, where I will commit the patch after short stay in mainline. The reason why the bug is latent on 4.7 seems to be the fac

Re: [C++ Patch] PR 52487

It's ill-formed to have a field with incomplete type. The best thing would be to complain about that before we get to literal_type_p so that errorcount is set, if that's not too complicated. Jason

[testsuite,avr,backport,4.7]: Backport some address-space tests from trunk

http://gcc.gnu.org/viewcvs?view=revision&revision=185696 Backported these address-space tests from trunk.

[Patch]: unwind-ia64: avoid long type

Hi, currently unwind-ia64.[ch] (and fde-glibc.c/fde-vms.c) use 'unsigned long' for a 64 bit integer type. That doesn't work well with VMS because 'long' is a 32 bit type on this platform… So this patch replaces 'unsigned long' with 'unw_word' (and the very few 'long' with 'unw_sword') in thes

[PATCH ARM] Add ctzsi2 to libiberty

This patch is mainly to fix a regression on trunk with support for __ctzsi2 on cores that do not have the CLZ instruction (notably pre-v5 and cortex-m0). It's needed after a recent change to longlong.h which now defines count_trailing_zeros to use the GCC builtin: unfortunately that builtin is use

[AVR,backported,4.7]: PR52148, PR52461, PR52484, PR52488, PR52496, PR52499, PR52505, PR52506, PR52507, PR52508, PR52543

http://gcc.gnu.org/viewcvs?view=revision&revision=185697 Backported these from trunk to 4.7. Johann libgcc/ Backport from 2012-03-07 mainline r185033. PR target/52507 * config/avr/lib1funcs.S (__movmemx_hi): Fix loop label in RAM-part. Backport from 2012-03-07 m

Re: [PATCH, alpha]: Some further configure cleanups

On 03/22/12 05:49, Uros Bizjak wrote: > 2012-03-22 Uros Bizjak > > * config/alpha/alpha.h (ASM_OUTPUT_ALIGN): Move to elf.h > (ASM_OUTPUT_SKIP): Move to vms.h > (ASM_OUTPUT_LOCAL): Ditto. > (ASM_OUTPUT_COMMON): Remove. > * config/alpha/elf.h (ASM_OUTPUT_DEF_FROM_DE

Re: [C++ Patch] PR 49152

On Thu, 22 Mar 2012, Paolo Carlini wrote: Hi, On 03/22/2012 05:25 AM, Gabriel Dos Reis wrote: On Wed, Mar 21, 2012 at 7:22 PM, Paolo Carlini wrote: Hi, this diagnostic issue is about not even trying to print expressions in error messages involving operators, and print operand types instead

Re: [PATCH, alpha]: Use generic config/elfos.h headers

On 03/22/12 00:47, Uros Bizjak wrote: > No, we are still calling default_file_start in alpha_file_start, so we > have to signal that we don't want start file directive. In committed > patch, I have redefined T_A_F_S_F_D as false for this purpose: But false is the default: DEFHOOKPOD (asm_file_sta

Re: [wwwdocs] Fix some nits in GCC 4.7 changes.html

Gerald Pfeifer writes: > On Fri, 16 Mar 2012, Rainer Orth wrote: >> While reading through the GCC 4.7 changes.html, I noticed a couple of >> nits that this patch fixes. It's mostly wording and grammar changes (so >> I may well be wrong and would appreciate a native speaker having a look > > T

Re: [PATCH, alpha]: Use generic config/elfos.h headers

On Thu, Mar 22, 2012 at 5:24 PM, Richard Henderson wrote: > On 03/22/12 00:47, Uros Bizjak wrote: >> No, we are still calling default_file_start in alpha_file_start, so we >> have to signal that we don't want start file directive. In committed >> patch, I have redefined T_A_F_S_F_D as false for th

Re: [wwwdocs] Fix some nits in GCC 4.7 changes.html

Gerald Pfeifer writes: > On Fri, 16 Mar 2012, Rainer Orth wrote: >> While reading through the GCC 4.7 changes.html, I noticed a couple of >> nits that this patch fixes. It's mostly wording and grammar changes (so >> I may well be wrong and would appreciate a native speaker having a look > > T

Re: [PATCH, alpha]: Use generic config/elfos.h headers

On 03/22/12 09:49, Uros Bizjak wrote: > The #define is now brought in form the generic elfos.h. We need to Ah, right. Ok. r~

Re: [C++ Patch] PR 49152

On Thu, Mar 22, 2012 at 11:13 AM, Marc Glisse wrote: > I haven't followed the whole diagnostic discussion, but what about printing > both the reconstructed expression and the types? Printing both isn't really the issue -- and we probably should. (And I thought we did in some cases.) What is at

Re: [C++ Patch] PR 52487

On 03/22/2012 03:49 PM, Jason Merrill wrote: It's ill-formed to have a field with incomplete type. The best thing would be to complain about that before we get to literal_type_p so that errorcount is set, if that's not too complicated. Agreed. The problem is that if we just change check_field_d

Re: [C++ Patch] PR 52487

On Thu, Mar 22, 2012 at 11:58 AM, Paolo Carlini wrote: > On 03/22/2012 03:49 PM, Jason Merrill wrote: >> >> It's ill-formed to have a field with incomplete type.  The best thing >> would be to complain about that before we get to literal_type_p so that >> errorcount is set, if that's not too compl

[patch, ARM] Implement Neon-optimized version of search_line_fast

search_line_fast in the lexer is documented as being a critical path. Several architectures already provide an optimized implementation of this routine using vector operations. This patch provides a version for ARM using the Neon vector extensions. Although it has not been tested on big-endian (I

Re: [Patch]: unwind-ia64: avoid long type

On 03/22/12 07:55, Tristan Gingold wrote: > 2012-03-22 Tristan Gingold > > * config/ia64/unwind-ia64.h: Declare unw_word and unw_sword. > (unw_table_entry): Use unw_word instead of unsigned long. > (_Unwind_FindTableEntry): Likewise. > * config/ia64/fde-glibc.c (_Unwind_

Re: [patch, ARM] Implement Neon-optimized version of search_line_fast

On 03/22/12 10:33, Richard Earnshaw wrote: > * lex.c (search_line_fast): Provide Neon-optimized version for ARM. Looks good to me. r~

[wwwdocs] Add link to Git mirror on main page

From a thread on IRC. We don't have a prominent link to the Git mirror and it's becoming one of the favourite ways for people to access the repo. Gerald, OK for the main page? I can't seem to be able to preview this change locally (dunno how the style.mhtml file gets loaded). Thanks. Di

Re: [C++ Patch] PR 52487

On 03/22/2012 12:58 PM, Paolo Carlini wrote: Anyway, I also think not calling literal_type_p from check_field_decls if the type is incomplete is pretty ugly, but I'm not sure which is the best way to make progress I guess that's OK. Just add a comment explaining that we'll get an error later.

Re: remove wrong code in immed_double_const

On Mar 22, 2012, at 6:15 AM, Michael Matz wrote: > That certainly is strictly better than any of the other possibilities, I > just didn't get the impression from your second mail to this thread that > you were even considering doing that. Good I was wrong. All I wanted, was to remove the assert

Re: remove wrong code in immed_double_const

On Mar 22, 2012, at 7:12 AM, Michael Matz wrote: > I see that you didn't remove the assert as part of this patch. I'll include that in my next patch. > I'd like to see what you like to do to this routine once the rest goes in. > In > particular I don't think just removing the assert will be e

[committed] Fix up __builtin_adjust_template handling in omp parallel/task code (PR middle-end/52547)

Hi! This patch fixes ICE when a nested function is called from within #pragma omp {parallel,task} region. We need to put the temporaries inside of the parallel/task region body, instead of the function body, as nothing is adding private clauses for them. Bootstrapped/regtested on x86_64-linux an

[C++ PATCH] Fix ICE on misplaced attribute (PR c++/52671)

Hi! The recent change to move the warning about misplaced attributes resulted in ICE on the following testcase. The problem is that the ENUMERAL_TYPE doesn't have type lang specific initialized and so CLASSTYPE_TEMPLATE_INSTANTIATION macro segfaults. Fixed thusly, bootstrapped/regtested on x86_64

Re: [C++ PATCH] Fix ICE on misplaced attribute (PR c++/52671)

OK. Jason

Re: remove wrong code in immed_double_const

On Mar 22, 2012, at 3:16 AM, Richard Sandiford wrote: > Sorry, meant we should leave the svn version as it is. Ah, I see now, I agree, I removed that bit (extend in the floating point case) of my change from the patch. > I think this should be s/is smaller than/is different from/. Sounds good,

Re: [committed] Fix up __builtin_adjust_template handling in omp parallel/task code (PR middle-end/52547)

On 03/22/12 12:34, Jakub Jelinek wrote: > 2012-03-22 Jakub Jelinek > > PR middle-end/52547 > * tree-nested.c (convert_tramp_reference_stmt): Call declare_vars > on any new_local_var_chain vars declared during recursing on > GIMPLE_OMP_PARALLEL or GIMPLE_OMP_TASK body. >

Re: [Patch V2] libgfortran: do not assume libm

Paolo Bonzini wrote: Il 22/03/2012 09:30, Tristan Gingold ha scritto: Bootstrapped without gfortran regressions on x86_64-darwin. Ok for trunk ? (I will submit a follow-up change in libquadmath once this change is approved). config/ 2012-03-22 Tristan Gingold * math.m4: New file. li

[pph] preliminary build support for stdc++.h.pph

Seems like -fpph doesn't work on the pph branch anymore, and instead -fpph-gen/-fpph-map/-fpph-hdr are used. The thing is, how are these flags used? unhelpful wiki page: http://gcc.gnu.org/wiki/pph Based on invoke.texi, I'm expecting something like this: 1. generate g++ -fpph-gen all.h; // g

Re: [fixincludes] Fix pthread.h failure (PR other/52626)

Rather simple... In a file called fixinclude.sum, put PASS: unique string or FAIL: unique string one per line, Bonus points if you can total passes and failures: I think you just put it into the check template so it creates the result. Since multiple fixes tweak the same file, ther

Re: [pph] preliminary build support for stdc++.h.pph

On 3/22/12, Benjamin De Kosnik wrote: > Seems like -fpph doesn't work on the pph branch anymore, and > instead -fpph-gen/-fpph-map/-fpph-hdr are used. > > The thing is, how are these flags used? -fpph-map=basename Enable reading PPH files, and instead of reading #include "basename.h", read

[wwdocs] update libstdc++ docs, links

Small filename tweaks to gcc's docs page for the generated 4.7.0 libstdc++ docs. -benjamin2012-03-22 Benjamin Kosnik * htdocs/onlinedocs/index.html: Adjust libstdc++ xml file names. Index: htdocs/onlinedocs/index.html ==

[pph] Distinguish template type parameters. (issue5881052)

This patch addresses the problem of two template instantiations appearing to be identical. They differ in TEMPLATE_PARM_NUM_SIBLINGS, and so that information needs to be pushed out to the template merge name. Test x1tmpldfltparm.cc is now passing the assert, though it has unneeded functions emitt

[v3] doc xml output names

Simple rename for XML filenames. The "-single" can be omitted. tested x86/linux -benjamin2012-03-22 Benjamin Kosnik * doc/Makefile.am: XML output names have only .xml suffix. * doc/Makefile.in: Regenerate. diff --git a/libstdc++-v3/doc/Makefile.am b/libstdc++-v3/doc/Makefile.am index 0f7b

Re: [fixincludes] Fix pthread.h failure (PR other/52626)

On Mar 22, 2012, at 3:40 PM, Bruce Korb wrote: > You can only figure out what has failed. Life goes on > Do I get the bonus points? make check and cat fixincludes.sum is necessary to get the bonus points. :-) Looks fine from what I can see.

[tpf] update for latest build system

Per request from IBM... * config/s390/s390.h (LINK_SPEC): Remove, no longer needed. (LIBSTDCXX): Change to CPP2. Index: config/s390/tpf.h === --- config/s390/tpf.h (revision 185677) +++ config/s390/tpf.h (working

Re: [C++ Patch] PR 52487

On 03/22/2012 07:28 PM, Jason Merrill wrote: On 03/22/2012 12:58 PM, Paolo Carlini wrote: Anyway, I also think not calling literal_type_p from check_field_decls if the type is incomplete is pretty ugly, but I'm not sure which is the best way to make progress I guess that's OK. Just add a commen

Re: [PATCH] Bug fix in store_bit_field_1 for big endian targets (issue 51893)

On Wed, Mar 21, 2012 at 10:53:07PM +0100, Eric Botcazou wrote: > > This patch (for 4.6) fixes a wrong subword index computation in > > store_bit_field_1 for big endian targets when value is at least 4 times > > bigger than a word (DI REG value with HI words). > > > > It fixes a regression on gcc.c-