Re: PATCH: PR rtl-optimization/54157: [x32] -maddress-mode=long failures
For the record, I can't approve this, but...
"H.J. Lu" writes:
> i386,md has
>
> (define_expand "extzv"
> [(set (match_operand:SI 0 "register_operand")
> (zero_extract:SI (match_operand 1 "ext_register_operand")
> (match_operand:SI 2 "const8_operand")
> (match_operand:SI 3 "const8_operand")))]
> ""
>
> and mode_for_extraction picks word_mode for operand 1 since
> its mode is VOIDmode. This patch changes mode_for_extraction
> to return the mode of operand 1 if the pattern accepts any mode.
> I added *jcc_btsi_mask_2 since combine now tries a different
> pattern, which leads to test failures on gcc.target/i386/bt-mask-1.c
> and gcc.target/i386/bt-mask-2. I didn't update *jcc_btsi_mask_1
> instead since I am not sure if it is used elsewhere. Tested on
> Linux/x86-64 and Linux/x32. OK for trunk?
the mode of the extraction operand is defined to be word_mode
for registers (see md.texi), so that at least would need to
be updated. But I'm not convinced that the wanted_inner_mode here:
if (! in_dest && unsignedp
- && mode_for_extraction (EP_extzv, -1) != MAX_MACHINE_MODE)
+ && mode_for_extraction (EP_extzv, -1, VOIDmode) != MAX_MACHINE_MODE)
{
- wanted_inner_reg_mode = mode_for_extraction (EP_extzv, 1);
- pos_mode = mode_for_extraction (EP_extzv, 3);
- extraction_mode = mode_for_extraction (EP_extzv, 0);
+ wanted_inner_reg_mode = mode_for_extraction (EP_extzv, 1,
+ inner_mode);
+ pos_mode = mode_for_extraction (EP_extzv, 3, VOIDmode);
+ extraction_mode = mode_for_extraction (EP_extzv, 0, VOIDmode);
}
is right. inner_mode is the mode of the thing we're extracting,
which doesn't ncessarily have anything to do with what the ext*
patterns support.
FWIW, in reply to your force_to_mode message, gen_lowpart_for_combine
looks a bit odd:
if (omode == imode)
return x;
/* Return identity if this is a CONST or symbolic reference. */
if (omode == Pmode
&& (GET_CODE (x) == CONST
|| GET_CODE (x) == SYMBOL_REF
|| GET_CODE (x) == LABEL_REF))
return x;
So if we know the modes are different, we nevertheless return the
original mode for CONST, SYMBOL_REF or LABEL_REF. Surely the caller
isn't going to be expecting that? If we're not prepared to change
the mode to the one that the caller asked for then I think we should
goto fail instead.
I don't know if you're hitting that or not.
Richard
RE: [PATCH, gcc/doc]: Document AMD btver2 enablement
Hi Gerald, Thank you. I committed changes as per your review comments. Ref: http://gcc.gnu.org/viewcvs?view=revision&revision=190151 Regards, Venkat. > -Original Message- > From: Gerald Pfeifer [mailto:ger...@pfeifer.com] > Sent: Sunday, August 05, 2012 12:24 AM > To: Kumar, Venkataramanan > Cc: gcc-patches@gcc.gnu.org > Subject: Re: [PATCH, gcc/doc]: Document AMD btver2 enablement > > On Sat, 4 Aug 2012, venkataramanan.ku...@amd.com wrote: > > Index: gcc/doc/extend.texi > > === > > +@item btver1 > > +AMD family 14h cpu. > > CPU... > > > @item amdfam15h > > AMD family 15h CPU. > > ...like you already had here. :-) > > > +@item btver2 > > +AMD family 16h cpu. > > CPU. > > > +@item btver2 > > +CPUs based on AMD Family 16h cores with x86-64 instruction set support. > (This > > +supersets MOVBE, F16C, BMI, AVX, PCL_MUL, AES, SSE4.2, SSE4.1, CX16, ABM, > > +SSE4A, SSSE3, SSE3, SSE2, SSE, MMX and 64-bit instruction set extensions.) > > I could not find "supsersets" as a verb in my dictionary: how > about "includes"? > > And I'd omit the parentheses around the second sentence, but this is > just a preference on my side; feel free to keep them if you prefer. > > Okay with these changes. > > Thanks, > Gerald
Re: [libiberty] add obstack macros (was Re: PR #53525 - track-macro-expansion performance regression)
On Sat, 4 Aug 2012, Ian Lance Taylor wrote: On Fri, 3 Aug 2012, Ian Lance Taylor wrote: I'm not sure where you are looking. I only see one call to _obstack_begin in the gcc directory, and it could easily be replaced with a call to obstack_specify_allocation instead. In libcpp/ mostly, but only 4 cases so that's not too many, I can change those with obstack_init/begin. +#define XOBFINISH(O, PT) ((PT) obstack_finish ((O))) For XOBNEW, etc., we use (T *) rather than (PT). Using (PT) seems error-probe--it's the only use of the obstack with a different type parameter. Why not use T rather than PT here, and return (T *)? I'd have to change many (about 60) occurences of XOBFINISH if I change that. I'd go for it if I was sure it's what we want, it can be a separate patch later on. I'm sorry to ask you to change a lot of code, but it simply doesn't make sense to me to have all but one macro take the type as an argument, and have one macro take a pointer to the type. They really have to be consistent. No problem, if anyone else doesn't object I'll change those (in a second patch, right?). Thanks, Dimitris
[Patch, Fortran] PR 35831: checking for dummy procedures
Hi all, here is a patch for a rather old PR, which deals with correctness checking for several cases, such as: 1) dummy procedures 2) proc-ptr assignments 3) type-bound procedure overloading The patch adds a new function 'check_result_characteristics' to do various checks on function results. This is largely analogous to 'check_dummy_characteristics'. In both of them, there are still a few attributes left to check, which I may add in a follow-up patch. Also I had to disable the warning for cases where we can not finally determine whether the string length or shape expressions are equal. The treatment for those cases should be improved at some point, or one should think about re-enabling the warnings. Regtested on x86_64-unknown-linux-gnu. Ok for trunk? Cheers, Janus 2012-08-05 Janus Weil PR fortran/35831 * interface.c (check_result_characteristics): New function, which checks the characteristics of function results. (gfc_compare_interfaces,gfc_check_typebound_override): Call it. 2012-08-05 Janus Weil PR fortran/35831 * gfortran.dg/dummy_procedure_5.f90: Modified. * gfortran.dg/dummy_procedure_8.f90: New. * gfortran.dg/interface_26.f90: Modified. * gfortran.dg/proc_ptr_11.f90: Modified. * gfortran.dg/proc_ptr_15.f90: Modified. * gfortran.dg/proc_ptr_result_5.f90: Modified. * gfortran.dg/typebound_override_1.f90: Modified. * gfortran.dg/typebound_proc_6.f03: Modified. pr35831.diff Description: Binary data dummy_procedure_8.f90 Description: Binary data
Re: [C++ Patch] PR 54165
OK. Jason
Re: [C++ Patch] PR 54161
On reflection, I think I prefer the status quo. As long as the pedwarn is on by default, I don't mind letting people disable it with -Wno-pointer-arith. Jason
[PATCH] Remove introduction of undefined overflow in emit_case_bit_test.
Richard,
the code in emit_case_bit_tests currently introduces a MINUS_EXPR in signed type
(without checking if signed types wrap or not), which could mean introduction of
undefined overflow.
This patch fixes this problem by performing the MINUS_EXPR in an unsigned type.
Doing so also enables the vrp optimization for the test-case.
Bootstrapped and reg-tested (ada inclusive) on x86_64.
OK for trunk?
Thanks,
- Tom
2012-08-06 Tom de Vries
* tree-switch-conversion.c (emit_case_bit_tests): Generate MINUS_EXPR in
unsigned type.
* gcc.dg/tree-ssa/vrp78.c: New test.
Index: gcc/tree-switch-conversion.c
===
--- gcc/tree-switch-conversion.c (revision 190139)
+++ gcc/tree-switch-conversion.c (working copy)
@@ -384,10 +384,10 @@ emit_case_bit_tests (gimple swtch, tree
gsi = gsi_last_bb (switch_bb);
- /* idx = (unsigned) (x - minval) */
- idx = fold_build2 (MINUS_EXPR, index_type, index_expr,
- fold_convert (index_type, minval));
- idx = fold_convert (unsigned_index_type, idx);
+ /* idx = (unsigned)x - minval */
+ idx = fold_convert (unsigned_index_type, index_expr);
+ idx = fold_build2 (MINUS_EXPR, unsigned_index_type, idx,
+ fold_convert (unsigned_index_type, minval));
idx = force_gimple_operand_gsi (&gsi, idx,
/*simple=*/true, NULL_TREE,
/*before=*/true, GSI_SAME_STMT);
Index: gcc/testsuite/gcc.dg/tree-ssa/vrp78.c
===
--- /dev/null (new file)
+++ gcc/testsuite/gcc.dg/tree-ssa/vrp78.c (revision 0)
@@ -0,0 +1,34 @@
+/* { dg-do link } */
+/* { dg-options "-O2" } */
+
+/* Based on f3 from vrp63.c, but with switch instead of if-chain. */
+
+extern void link_error (void);
+
+void
+f3 (int s)
+{
+ if (s >> 3 == -2)
+/* s in range [ -16, -9]. */
+;
+ else
+{
+ /* s in range ~[-16, -9], so none of the case labels can be taken. */
+ switch (s)
+ {
+ case -16:
+ case -12:
+ case -9:
+ link_error ();
+ break;
+ default:
+ break;
+ }
+}
+}
+
+int
+main ()
+{
+ return 0;
+}
Re: [PATCH][2/2] Remove referenced vars
On Thu, Aug 2, 2012 at 1:12 AM, Richard Guenther wrote: > On Wed, 1 Aug 2012, Richard Guenther wrote: > >> >> This is the combined patch, ontop of 2a and 2b this removes all >> gimple_referenced_vars and add_referenced_vars calls. >> >> Bootstrapped on x86_64-unknown-linux-gnu, testing in progress. >> >> GCC is a better place w/o referenced vars ;) > > Re-bootstrapped and tested on x86_64-unknown-linux-gnu ontop of > what has been actually committed. Adjusted a few dump-scans in > points-to analysis. This will also result in > > FAIL: gcc.dg/pr52558-1.c scan-tree-dump-times lim1 "MEM count_lsm.. > count_lsm_fl > ag" 1 > FAIL: gcc.dg/pr52558-2.c scan-tree-dump-times lim1 "MEM.*g_2_lsm_flag" 1 > FAIL: gcc.dg/tm/reg-promotion.c scan-tree-dump-times lim1 "MEM count_lsm.. > count > _lsm_flag" 1 > > as I could not figure out what to replace the previously senseless > dump scanning with. There are no changelog entries for the addition > of these testcases, so I'm not sure whom to nag. gcc.dg/pr52558-1.c > was committed by Aldyh at least. Aldyh - these tests scan for random > strings and in no way guarantee that the store remains to be executed > conditional only. Maybe you should amend the lim dump in case you > want to only ensure that the lim transform uses conditional replacement? > > Committed to trunk. > > Richard. > > 2012-08-02 Richard Guenther > > * tree-dfa.c (referenced_var_lookup): Remove. > (find_referenced_vars): Remove. > (pass_referenced_vars): Likewise. > (make_rename_temp): Do not add referenced vars. > (dump_referenced_vars): Remove. > (debug_referenced_vars): Likewise. > (dump_dfa_stats): Do not dump referenced var stats. > (find_vars_r): Remove. > (find_referenced_vars_in): Likewise. > (referenced_var_check_and_insert): Likewise. > (add_referenced_var_1): Likewise. > (remove_referenced_var): Likewise. > * tree-flow.h (referenced_var_lookup): Likewise. > (struct gimple_df): Remove referenced_vars member. > (typedef referenced_var_iterator): Remove. > (FOR_EACH_REFERENCED_VAR): Likewise. > (num_referenced_vars): Likewise. > (debug_referenced_vars, dump_referenced_vars): Likewise. > (add_referenced_var_1, add_referenced_var): Likewise. > (remove_referenced_var): Likewise. > (find_referenced_vars_in): Likewise. > * tree-flow-inline.h (gimple_referenced_vars): Remove. > (first_referenced_var): Likewise. > (end_referenced_vars_p): Likewise. > (next_referenced_var): Likewise. > * cfgexpand.c (update_alias_info_with_stack_vars): Remove assert. > * gimple-pretty-print.c (pp_points_to_solution): Dump UIDs > unconditionally. > * tree-into-ssa.c (dump_decl_set): Likewise. > (pass_build_ssa): Do not require PROP_referenced_vars. > * tree-ssa.c (target_for_debug_bind): Virtual operands are > not suitable, but all register type vars are. > (init_tree_ssa): Do not allocate referenced vars. > (delete_tree_ssa): Do not free referenced vars. > * cgraphunit.c (init_lowered_empty_function): Do not set > PROP_referenced_vars. > (assemble_thunk): Do not add referenced vars. > * gimple-fold.c (canonicalize_constructor_val): Likewise. > (gimplify_and_update_call_from_tree): Likewise. > * gimple-streamer-in.c (input_bb): Likewise. > * passes.c (init_optimization_passes): Do not run > pass_referenced_vars. > (dump_properties): Do not dump PROP_referenced_vars. > * tree-inline.c (remap_decl): Do not add referenced vars. > (remap_gimple_op_r): Likewise. > (copy_tree_body_r): Likewise. > (copy_bb): Likewise. > (setup_one_parameter): Likewise. > (declare_return_variable): Likewise. > (copy_decl_for_dup_finish): Likewise. > (copy_arguments_for_versioning): Likewise. > (tree_function_versioning): Likewise. > * tree-pass.h (PROP_referenced_vars): Remove. > (pass_referenced_vars): Likewise. > * tree-profile.c (gimple_gen_edge_profiler): Do not add referenced > vars. > (gimple_gen_interval_profiler): Likewise. > (gimple_gen_pow2_profiler): Likewise. > (gimple_gen_one_value_profiler): Likewise. > (gimple_gen_ic_profiler): Likewise. > (gimple_gen_average_profiler): Likewise. > (gimple_gen_ior_profiler): Likewise. > * tree-ssa-live.c (remove_unused_locals): Do not touch referenced > vars. > * tree-cfg.c (replace_ssa_name): Do not add referenced vars. > (move_stmt_op): Likewise. > * tree-ssa-operands.c (create_vop_var): Likewise. > * gimple-low.c (record_vars_into): Likewise. > * gimplify.c (gimple_regimplify_operands): Likewise. > (force_gimple_operand_1): Likewise. > * omp-low.
Re: [C++ Patch] PR 54161
On 08/06/2012 01:29 AM, Jason Merrill wrote: On reflection, I think I prefer the status quo. As long as the pedwarn is on by default, I don't mind letting people disable it with -Wno-pointer-arith. I hear you. Then, let's just close the PR. Thanks! Paolo.
Re: [libiberty] add obstack macros (was Re: PR #53525 - track-macro-expansion performance regression)
On Sun, Aug 5, 2012 at 11:44 AM, Dimitrios Apostolou wrote: > > No problem, if anyone else doesn't object I'll change those (in a second > patch, right?). Can you please resend the patch you propose to check into mainline? Thanks. Ian
[PATCH] put exception tables for comdat functions in comdat, too
For C++ functions that are emitted to a comdat section (template
instantiations and the like), the corresponding exception handling
information is presently being emitted to the non-comdat
.gcc_except_table section. This means that you can end up with multiple
copies of the exception tables in the executable from different objects,
even though the linker can remove all but one copy of the function it
goes with.
We do have a test case for this, but I don't know how to mash it into
the Dejagnu framework. My hand-testing involved comparing the
before/after output of readelf to see the new exception table sections
in the group, and verifying that the size of the executable is smaller
with the patch than without it.
Otherwise, I bootstrapped and regression tested on x86_64, and also
tested on mipsisa32r2-sde-elf since I had an existing build tree for
that set up. This patch has also been present in our local source base
for a while so it's been well-tested in that context. OK for mainline?
-Sandra
2012-08-04 Paul Brook
Sandra Loosemore
gcc/
* except.c (switch_to_exception_section): Place tables for
DECL_ONE_ONLY functions in comdat groups.
Index: gcc/except.c
===
--- gcc/except.c (revision 190149)
+++ gcc/except.c (working copy)
@@ -2777,11 +2777,16 @@ switch_to_exception_section (const char
flags = SECTION_WRITE;
#ifdef HAVE_LD_EH_GC_SECTIONS
- if (flag_function_sections)
+ if (flag_function_sections
+ || (DECL_ONE_ONLY (current_function_decl) && HAVE_COMDAT_GROUP))
{
char *section_name = XNEWVEC (char, strlen (fnname) + 32);
+ /* The EH table must match the code section, so only mark
+ it linkonce if we have COMDAT groups to tie them together. */
+ if (DECL_ONE_ONLY (current_function_decl) && HAVE_COMDAT_GROUP)
+ flags |= SECTION_LINKONCE;
sprintf (section_name, ".gcc_except_table.%s", fnname);
- s = get_section (section_name, flags, NULL);
+ s = get_section (section_name, flags, current_function_decl);
free (section_name);
}
else
Re: [PATCH] Improve ifcombine (PR 52005)
Hello, do you have an opinion on this patch (available here: http://gcc.gnu.org/ml/gcc-patches/2012-07/msg01352.html ) ? Or should we go back to my old patch, or Andrew's patch? On Thu, 26 Jul 2012, Marc Glisse wrote: Hello, here is a new version of the ifcombine patch, which handles Andrew's examples (though only with -O2 for one of them, same_phi_args_p returns false otherwise). Note that the new patch never calls maybe_fold_or_comparisons, only maybe_fold_and_comparisons. It would be possible to call maybe_fold_or_comparisons, when maybe_fold_and_comparisons fails, but I don't know if there are optimizations that one does and not the other, except for the odd trapping-math case. Bootstrap and testsuite are fine. 2012-07-26 Marc Glisse gcc/ PR tree-optimization/51938 PR tree-optimization/52005 * tree-ssa-ifcombine.c (ifcombine_ifandif): New parameters for inverted conditions. (ifcombine_iforif): Remove, merge code into ifcombine_ifandif. (tree_ssa_ifcombine_bb): Update calls to the above. Detect !a&&b and !a||b patterns. gcc/testsuite/ PR tree-optimization/51938 PR tree-optimization/52005 * gcc.dg/tree-ssa/ssa-ifcombine-8.c: New testcase. * gcc.dg/tree-ssa/ssa-ifcombine-9.c: Likewise. * gcc.dg/tree-ssa/ssa-ifcombine-10.c: Likewise. * gcc.dg/tree-ssa/ssa-ifcombine-11.c: Likewise. -- Marc Glisse
[patch] speed up ifcvt:cond_move_convert_if_block
Hello, In PR54146, ifcvt spends a lot of time just clearing memory. This patch changes the value maps to pointer-maps to fix this issue. Bootstrapped&tested on x86_64-unknown-linux-gnu. OK? Ciao! Steven PR54146_ifcvt.diff Description: Binary data
