diff mbox

Patch to fix PR79131

Message ID 49091620-2b61-fe51-d00f-f8e567bb5f62@redhat.com
State New
Headers show

Commit Message

Vladimir Makarov Jan. 26, 2017, 5:09 p.m. UTC
The following patch fixes

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79131

The patch also adapts IP IRA in LRA because without it GCC IP RA tests 
become broken (it was just a luck that the tests worked before the patch).

The patch was successfully bootstrapped and tested on x86-64.

Committed as rev. 244942.

Comments

Christophe Lyon Jan. 30, 2017, 8:59 p.m. UTC | #1
Hi Vladimir,

On 26 January 2017 at 18:09, Vladimir Makarov <vmakarov@redhat.com> wrote:
> The following patch fixes
>
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79131
>
> The patch also adapts IP IRA in LRA because without it GCC IP RA tests
> become broken (it was just a luck that the tests worked before the patch).
>
> The patch was successfully bootstrapped and tested on x86-64.
>
> Committed as rev. 244942.
>
>
After this patch, I've noticed regressions on arm:
FAIL:  gcc.target/arm/neon-for-64bits-1.c scan-assembler-times vshr 0
Your follow-up patch didn't fix this.
There are now 6 vshr instructions generated.

Can you check ?

Thanks,
Christophe
Kyrill Tkachov Jan. 31, 2017, 9:05 a.m. UTC | #2
Hi Christophe,

On 30/01/17 20:59, Christophe Lyon wrote:
> Hi Vladimir,
>
> On 26 January 2017 at 18:09, Vladimir Makarov <vmakarov@redhat.com> wrote:
>> The following patch fixes
>>
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79131
>>
>> The patch also adapts IP IRA in LRA because without it GCC IP RA tests
>> become broken (it was just a luck that the tests worked before the patch).
>>
>> The patch was successfully bootstrapped and tested on x86-64.
>>
>> Committed as rev. 244942.
>>
>>
> After this patch, I've noticed regressions on arm:
> FAIL:  gcc.target/arm/neon-for-64bits-1.c scan-assembler-times vshr 0
> Your follow-up patch didn't fix this.
> There are now 6 vshr instructions generated.
>
> Can you check ?

I've opened PR 79282.

I think it may be an issue exposed by Vladimir's patch.

Kyrill

> Thanks,
> Christophe
Christophe Lyon Jan. 31, 2017, 9:39 a.m. UTC | #3
On 31 January 2017 at 10:05, Kyrill Tkachov <kyrylo.tkachov@foss.arm.com> wrote:
> Hi Christophe,
>
> On 30/01/17 20:59, Christophe Lyon wrote:
>>
>> Hi Vladimir,
>>
>> On 26 January 2017 at 18:09, Vladimir Makarov <vmakarov@redhat.com> wrote:
>>>
>>> The following patch fixes
>>>
>>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79131
>>>
>>> The patch also adapts IP IRA in LRA because without it GCC IP RA tests
>>> become broken (it was just a luck that the tests worked before the
>>> patch).
>>>
>>> The patch was successfully bootstrapped and tested on x86-64.
>>>
>>> Committed as rev. 244942.
>>>
>>>
>> After this patch, I've noticed regressions on arm:
>> FAIL:  gcc.target/arm/neon-for-64bits-1.c scan-assembler-times vshr 0
>> Your follow-up patch didn't fix this.
>> There are now 6 vshr instructions generated.
>>
>> Can you check ?
>
>
> I've opened PR 79282.
>
> I think it may be an issue exposed by Vladimir's patch.

That's my feeling too.
Thanks for the heads-up, for some reason I did not receive an email
from bugzilla.

>
> Kyrill
>
>> Thanks,
>> Christophe
>
>
diff mbox

Patch

Index: ChangeLog
===================================================================
--- ChangeLog	(revision 244937)
+++ ChangeLog	(working copy)
@@ -1,3 +1,16 @@ 
+2017-01-26  Vladimir Makarov  <vmakarov@redhat.com>
+
+	PR target/79131
+	* lra-assigns.c (setup_live_pseudos_and_spill_after_risky): Take
+	endianess for subregs into account.
+	* lra-constraints.c (lra_constraints): Do risky transformations
+	always on the first iteration.
+	* lra-lives.c (check_pseudos_live_through_calls): Add arg
+	last_call_used_reg_set.
+	(process_bb_lives): Define and use last_call_used_reg_set.
+	* lra.c (lra): Always continue after lra_constraints on the first
+	iteration.
+
 2017-01-26  Jakub Jelinek  <jakub@redhat.com>
 
 	* config/i386/avx512fintrin.h (_ktest_mask16_u8,
Index: lra-assigns.c
===================================================================
--- lra-assigns.c	(revision 244595)
+++ lra-assigns.c	(working copy)
@@ -1179,9 +1179,19 @@  setup_live_pseudos_and_spill_after_risky
 	    /* If it is multi-register pseudos they should start on
 	       the same hard register.	*/
 	    || hard_regno != reg_renumber[conflict_regno])
-	  add_to_hard_reg_set (&conflict_set,
-			       lra_reg_info[conflict_regno].biggest_mode,
-			       reg_renumber[conflict_regno]);
+	  {
+	    int conflict_hard_regno = reg_renumber[conflict_regno];
+	    machine_mode biggest_mode = lra_reg_info[conflict_regno].biggest_mode;
+	    int biggest_nregs = hard_regno_nregs[conflict_hard_regno][biggest_mode];
+	    int nregs_diff = (biggest_nregs
+			      - (hard_regno_nregs
+				 [conflict_hard_regno]
+				 [PSEUDO_REGNO_MODE (conflict_regno)]));
+	    add_to_hard_reg_set (&conflict_set,
+				 biggest_mode,
+				 conflict_hard_regno
+				 - (WORDS_BIG_ENDIAN ? nregs_diff : 0));
+	  }
       if (! overlaps_hard_reg_set_p (conflict_set, mode, hard_regno))
 	{
 	  update_lives (regno, false);
Index: lra-constraints.c
===================================================================
--- lra-constraints.c	(revision 244595)
+++ lra-constraints.c	(working copy)
@@ -4511,7 +4511,10 @@  lra_constraints (bool first_p)
       && REGNO (pic_offset_table_rtx) >= FIRST_PSEUDO_REGISTER)
     lra_risky_transformations_p = true;
   else
-    lra_risky_transformations_p = false;
+    /* On the first iteration we should check IRA assignment
+       correctness.  In rare cases, the assignments can be wrong as
+       early clobbers operands are ignored in IRA.  */
+    lra_risky_transformations_p = first_p;
   new_insn_uid_start = get_max_uid ();
   new_regno_start = first_p ? lra_constraint_new_regno_start : max_reg_num ();
   /* Mark used hard regs for target stack size calulations.  */
Index: lra-lives.c
===================================================================
--- lra-lives.c	(revision 244595)
+++ lra-lives.c	(working copy)
@@ -560,10 +560,11 @@  lra_setup_reload_pseudo_preferenced_hard
 }
 
 /* Check that REGNO living through calls and setjumps, set up conflict
-   regs, and clear corresponding bits in PSEUDOS_LIVE_THROUGH_CALLS and
-   PSEUDOS_LIVE_THROUGH_SETJUMPS.  */
+   regs using LAST_CALL_USED_REG_SET, and clear corresponding bits in
+   PSEUDOS_LIVE_THROUGH_CALLS and PSEUDOS_LIVE_THROUGH_SETJUMPS.  */
 static inline void
-check_pseudos_live_through_calls (int regno)
+check_pseudos_live_through_calls (int regno,
+				  HARD_REG_SET last_call_used_reg_set)
 {
   int hr;
 
@@ -571,7 +572,7 @@  check_pseudos_live_through_calls (int re
     return;
   sparseset_clear_bit (pseudos_live_through_calls, regno);
   IOR_HARD_REG_SET (lra_reg_info[regno].conflict_hard_regs,
-		    call_used_reg_set);
+		    last_call_used_reg_set);
 
   for (hr = 0; hr < FIRST_PSEUDO_REGISTER; hr++)
     if (HARD_REGNO_CALL_PART_CLOBBERED (hr, PSEUDO_REGNO_MODE (regno)))
@@ -604,11 +605,13 @@  process_bb_lives (basic_block bb, int &c
   rtx_insn *next;
   rtx link, *link_loc;
   bool need_curr_point_incr;
-
+  HARD_REG_SET last_call_used_reg_set;
+  
   reg_live_out = df_get_live_out (bb);
   sparseset_clear (pseudos_live);
   sparseset_clear (pseudos_live_through_calls);
   sparseset_clear (pseudos_live_through_setjumps);
+  CLEAR_HARD_REG_SET (last_call_used_reg_set);
   REG_SET_TO_HARD_REG_SET (hard_regs_live, reg_live_out);
   AND_COMPL_HARD_REG_SET (hard_regs_live, eliminable_regset);
   EXECUTE_IF_SET_IN_BITMAP (reg_live_out, FIRST_PSEUDO_REGISTER, j, bi)
@@ -795,7 +798,8 @@  process_bb_lives (basic_block bb, int &c
 	    need_curr_point_incr
 	      |= mark_regno_live (reg->regno, reg->biggest_mode,
 				  curr_point);
-	    check_pseudos_live_through_calls (reg->regno);
+	    check_pseudos_live_through_calls (reg->regno,
+					      last_call_used_reg_set);
 	  }
 
       for (reg = curr_static_id->hard_regs; reg != NULL; reg = reg->next)
@@ -831,15 +835,27 @@  process_bb_lives (basic_block bb, int &c
 
       if (call_p)
 	{
-	  if (flag_ipa_ra)
+	  if (! flag_ipa_ra)
+	    COPY_HARD_REG_SET(last_call_used_reg_set, call_used_reg_set);
+	  else
 	    {
 	      HARD_REG_SET this_call_used_reg_set;
 	      get_call_reg_set_usage (curr_insn, &this_call_used_reg_set,
 				      call_used_reg_set);
 
+	      bool flush = (! hard_reg_set_empty_p (last_call_used_reg_set)
+			    && ! hard_reg_set_equal_p (last_call_used_reg_set,
+						       this_call_used_reg_set));
+
 	      EXECUTE_IF_SET_IN_SPARSESET (pseudos_live, j)
-		IOR_HARD_REG_SET (lra_reg_info[j].actual_call_used_reg_set,
-				  this_call_used_reg_set);
+		{
+		  IOR_HARD_REG_SET (lra_reg_info[j].actual_call_used_reg_set,
+				    this_call_used_reg_set);
+		  if (flush)
+		    check_pseudos_live_through_calls
+		      (j, last_call_used_reg_set);
+		}
+	      COPY_HARD_REG_SET(last_call_used_reg_set, this_call_used_reg_set);
 	    }
 
 	  sparseset_ior (pseudos_live_through_calls,
@@ -866,7 +882,8 @@  process_bb_lives (basic_block bb, int &c
 	    need_curr_point_incr
 	      |= mark_regno_live (reg->regno, reg->biggest_mode,
 				  curr_point);
-	    check_pseudos_live_through_calls (reg->regno);
+	    check_pseudos_live_through_calls (reg->regno,
+					      last_call_used_reg_set);
 	  }
 
       for (reg = curr_static_id->hard_regs; reg != NULL; reg = reg->next)
@@ -1009,7 +1026,7 @@  process_bb_lives (basic_block bb, int &c
       if (sparseset_cardinality (pseudos_live_through_calls) == 0)
 	break;
       if (sparseset_bit_p (pseudos_live_through_calls, j))
-	check_pseudos_live_through_calls (j);
+	check_pseudos_live_through_calls (j, last_call_used_reg_set);
     }
 
   if (need_curr_point_incr)
Index: lra.c
===================================================================
--- lra.c	(revision 244595)
+++ lra.c	(working copy)
@@ -2295,7 +2295,7 @@  void
 lra (FILE *f)
 {
   int i;
-  bool live_p, scratch_p, inserted_p;
+  bool live_p, inserted_p;
 
   lra_dump_file = f;
 
@@ -2332,7 +2332,6 @@  lra (FILE *f)
   lra_constraint_new_regno_start = lra_new_regno_start = max_reg_num ();
   lra_bad_spill_regno_start = INT_MAX;
   remove_scratches ();
-  scratch_p = lra_constraint_new_regno_start != max_reg_num ();
 
   /* A function that has a non-local label that can reach the exit
      block via non-exceptional paths must save all call-saved
@@ -2372,11 +2371,12 @@  lra (FILE *f)
       for (;;)
 	{
 	  /* We should try to assign hard registers to scratches even
-	     if there were no RTL transformations in
-	     lra_constraints.  */
+	     if there were no RTL transformations in lra_constraints.
+	     Also we should check IRA assignments on the first
+	     iteration as they can be wrong because of early clobbers
+	     operands which are ignored in IRA.  */
 	  if (! lra_constraints (lra_constraint_iter == 0)
-	      && (lra_constraint_iter > 1
-		  || (! scratch_p && ! caller_save_needed)))
+	      && lra_constraint_iter > 1)
 	    break;
 	  /* Constraint transformations may result in that eliminable
 	     hard regs become uneliminable and pseudos which use them
Index: testsuite/ChangeLog
===================================================================
--- testsuite/ChangeLog	(revision 244937)
+++ testsuite/ChangeLog	(working copy)
@@ -1,3 +1,8 @@ 
+2017-01-26  Vladimir Makarov  <vmakarov@redhat.com>
+
+	PR target/79131
+	* gcc.target/arm/pr79131.c: New.
+
 2017-01-26  Bin Cheng  <bin.cheng@arm.com>
 
 	* gcc.target/aarch64/ldp_vec_64_1.c: Xfail.
Index: testsuite/gcc.target/arm/pr79131.c
===================================================================
--- testsuite/gcc.target/arm/pr79131.c	(revision 0)
+++ testsuite/gcc.target/arm/pr79131.c	(working copy)
@@ -0,0 +1,16 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O2 -mbig-endian" } */
+
+long long a;
+enum { NILFS_SEGMENT_USAGE_ACTIVE, NILFS_SEGMENT_USAGE_DIRTY } b;
+void nilfs_sufile_mod_counter(long long p1) {
+  long c = p1;
+  unsigned d = __builtin_bswap64(a);
+  a = __builtin_bswap64(d + c);
+}
+void nilfs_sufile_do_free() {
+  int e, f;
+  e = __builtin_bswap32(b) & 1 << NILFS_SEGMENT_USAGE_DIRTY;
+  f = e;
+  nilfs_sufile_mod_counter(f ? -1 : 0);
+}